Sensing Heroes and Villains: Character-Schema and the Disposition Formation Process

Method

In Study 1, we created two characters that were designed to vary in the extent to which they represented visual characteristics associated with hero and villain schema. We then randomly assigned participants (N = 261; n_Male = 147, 56.3%; n_Female = 114, 43.7%; n_{White/Caucasian} = 118, 45.2%; n_{Black/African American} = 19, 7.4%; n_Asian = 85, 32.6%; each remaining race accounted for less than 5% of the sample; M_Age = 20.02, SD_Age = 2.84) to view either the heroic character (henceforth, hero: n = 133) or the villainous character (henceforth, villain: n = 128) and rate him along the measures used in Eden et al.’s (2015) study. Random assignment appeared successful as we found no significant differences between the two conditions in terms of sex, χ²(df = 1) = 0.07, p = .79; ethnicity, χ²(df = 3) = 0.30, p = .96 ¹ ; or age, t(258) = 0.85, p = .40. Participants were recruited from communication courses at a large university in the northeastern United States and received course credit for participation. Data collection occurred from April 27 to May 1, 2015.

Character stimulus material

We utilized computer software to create and manipulate character-schema to maximize the internal validity of our conclusions. A single character was created through the character creation feature of the video game EVE Online (CCP Games, 2003). We then manipulated (a) facial expression, friendly for the hero versus stern for the villain; (b) hair color and outfit color, light brown for hero versus black for villain; and (c) facial attractiveness, light facial scarring which was present only for the villain (see Figure 1).

OPEN IN VIEWER

Figure 1.

Hero and villain character-schema stimuli.

Results and Discussion

We tested our hypotheses using a multivariate analysis of variance (MANOVA). The multivariate test indicated a significant effect of character-schema, Pillai’s trace = .40, F(10, 249) = 16.51, p < .001, η p 2 = .40. ² See Table 1 for descriptive statistics and univariate results.

OPEN IN VIEWER

Table 1.

Estimated Marginal Means and Univariate Results for Study 1.

	Hero character				Villain character				Univariate ANOVA testing difference between characters
	M	SD	95% CI		M	SD	95% CI
			LL	UL			LL	UL
Moral intuitions
Care	3.67	1.32	3.48	3.86	2.39	0.86	2.20	2.59	F = 84.70***	η p 2 = .25
Fairness	3.26	1.16	3.08	3.44	2.61	0.95	2.43	2.80	F = 23.95***	η p 2 = .09
Authority	4.09	1.30	3.88	4.31	2.88	1.24	2.66	3.10	F = 59.44***	η p 2 = .19
Loyalty	4.28	1.29	4.06	4.50	3.35	1.25	3.13	3.57	F = 34.66***	η p 2 = .12
Purity	4.13	1.27	3.92	4.34	2.95	1.21	2.73	3.17	F = 58.88***	η p 2 = .19
Person-perception
Warmth	3.79	1.05	3.62	3.96	2.63	0.96	2.45	2.80	F = 86.70***	η p 2 = .25
Duplicity	3.88	1.13	3.70	4.05	4.99	0.85	4.82	5.17	F = 81.16***	η p 2 = .24
Competence	5.39	0.82	5.24	5.54	5.30	0.94	5.15	5.46	F = 0.57, ns	η p 2 = .00
Heroic	4.22	1.39	3.99	4.45	3.01	1.25	2.78	3.24	F = 54.52***	η p 2 = .17
Villainous	4.48	1.21	4.29	4.68	5.52	1.07	5.32	5.72	F = 52.87***	η p 2 = .17

Note. df for univariate ANOVA are (1, 258). CI = confidence interval; LL = lower limit; UL = upper limit; ns = nonsignificant.

***

p ≤ .001.

Results were consistent with all the hypotheses. The hero character was rated as more moral than the villain character across all five moral intuitions, with moderate to large univariate effect sizes ( η p 2 : minimum = .09, maximum = .25, M = .17, SD = .07). Importantly, the 95% confidence intervals (CIs) did not overlap (see Table 1), suggesting that our two characters led to contrasting perceptions. The hero character was also rated as being warmer ( η p 2 = .25) and less duplicitous ( η p 2 = .24) than the villain character with nonoverlapping 95% CIs. The characters did not differ in competence; an equivalence test (Weber & Popova, 2012) was consistent with the null, given a delta = .30, t(258) = −0.76, p = .001. η p 2 = .00. Finally, the manipulation check items showed patterns consistent with expectations. The heroic character was perceived to be more heroic and less villainous than the villain character. Again, the 95% CIs did not overlap. Overall, the results of Study 1 suggest that visual cues can activate character-schema; our artificially created hero/villain stimuli perfectly replicated the patterns observed in Eden et al.’s (2015) study.

Method

Study 2 utilized a mixed design to test H4 and H5. Participants (N = 75; n_Male = 30, 40%; n_Female = 45, 60%; n_{White/Caucasian} = 29, 38.7%; n_{Black/African American} = 6, 8.0%; n_Asian = 24, 32.0%; n_{Hispanic/Latino} = 5, 6.6%; each remaining race accounted for less than 5% of the sample; M_Age = 19.83, SD_Age = 1.89) rated both the characters from Study 1 and were randomly assigned to the order of evaluation (order: viewing and rating the hero first, n = 37; or viewing and rating the villain first, n = 38, between-subjects). ³ Random assignment was confirmed with no significant effects of condition being found for sex, χ²(df = 1) = 0.72, p = .40; ethnicity, χ²(df = 3) = 4.43, p = .22; or age, t(73) = 1.80, p = .08. ⁴ Data collection occurred from September 21 to September 30, 2015. ⁵

Narrative

Academics from various fields have provided definitions of narrative, with the most basic being “a series of logically and chronologically related events” (Bal, 1997, p. 5). This basic definition, however, is somewhat incomplete; as noted by Rudrum (2005), a schematic describing furniture assembly would fit such a definition. Thus, narratologists generally agree that other components are necessary before a series of events qualifies as a narrative. Four components that are consistently considered necessary are characters, events, temporal progression, and narration (cf., Altman, 2008; Bal, 1997; Green & Brock, 2000). Characters are beings within a story that experience events (i.e., transitions “from one state to another state”; Bal, 1997, p. 5). Temporal progression and narration relate to the sequential (though not necessarily chronological) unfolding of the story—a process Altman (2008) described as “following.”

Narratives relate to the disposition formation processes described by Zillmann (2000) and Raney (2004) as follows (see also Lee & Shapiro, 2014; Sanders, 2005). Schema biases judgments prior to narrative exposure. Viewers then witness narrative events unfold. As the character engages in behavior, viewers may revise or reinforce their initial judgments. Thus, to test Raney’s propositions, the basic narrative format requires that (a) character-schema biases character judgments prior to the narrative, (b) the narrative describes an event in which the character behaves in a more or less moral manner, and (c) a resolution is present. Character evaluations made prior to the narrative can then be compared with those made afterward.

An Integrated Model

Raney stated that his schema-based hypothesis was “an extension, not a contradiction of disposition theory” (Raney, 2004, p. 356). Thus, it should be possible to create a conceptual model of Zillmann’s (2000) MSTDR that includes the schema-based relationships proposed by Raney (2004). We propose such a model in Figure 2. Zillmann’s (2000) MSTDR is represented in Paths 3 and 7. Behavior elicits approbation, which influences moral judgments of characters. The remaining paths represent Raney’s schema-based hypotheses. Path 1 represents character-schema-based effects on moral judgment prior to viewing any behavior. Path 2 represents Raney’s moral justification proposition. Although Raney’s schema-based propositions reflect a main effect, we included Path 4—which represents a potential interaction between character-schema and behavior on approbation—as a way to verify this claim. This path should be zero or positive due to mathematical necessity. The only findings that would be inconsistent with Raney’s claims would be a negative character-schema-behavior interaction, indicating a crossover effect whereby the hero received greater approbation than a villain for less moral behavior but less approbation than a villain for more moral behavior. Raney’s moral justification hypothesis is further represented in Paths 5 and 6. Path 5 should be positive with initial moral judgments, leading to more positive moral approbation of subsequent behaviors. Similarly, Path 6 should be positive with past judgments of a character carried over into future judgments.

OPEN IN VIEWER

Figure 2.

Conceptual path model illustrating Zillmann’s (2000) and Raney’s (2004) propositions.

Notably, our hypothesized model represents our rigid interpretation of Raney’s and Zillmann’s propositions. Both scholars built considerable flexibility into their models, and their approaches could be read as completely consistent with each other. For example, Zillmann (2000) included a feedback loop between affect and approbation, which he described as reflecting “the influence of formed dispositions on moral judgment” (p. 54). Similarly, Raney (2004) argued that narratives allow for moral judgment to alter dispositions in a manner that contradicts schema-based perceptions. Some may view our rigid interpretation as a weakness. However, without such an interpretation, both models become unfalsifiable. Thus, we have attempted to provide a template for both testing their predictions and discovering theoretical ambiguity.

Method

We conducted a 2 (character-schema: hero or villain) × 2 (behavior: more or less moral) between-subjects factorial experiment. Participants (N = 130; n_Male = 64, 49.2%; n_Female = 66, 50.8%; n_{White/Caucasian} = 58, 44.6%; n_{Black/African American} = 12, 9.2%; n_Asian = 38, 29.2%; each remaining race accounted for less than 5% of the sample; M_Age = 20.87, SD_Age = 3.98) were randomly assigned to the conditions (n_{Hero-More Moral} = 33; n_{Hero-Less Moral} = 32; n_{Villain-More Moral} = 32; n_{Villain-Less Moral} = 33). After evaluating their randomly assigned character on the CMFQ (Time 1; T1), participants read a story in which the character enacted a behavior that was either more or less morally acceptable. Participants then completed an approbation of behavior measure, followed by a re-evaluation of the character on the CMFQ (Time 2; T2). Random assignment was confirmed: condition and sex, χ²(df = 1) = 0.12, p = .73; ethnicity, χ²(df = 3) = 0.94, p = .82; and age, all Fs < 1. Data collection occurred from September 21 to September 30, 2015.

Results and Discussion

We tested the hypothesized path model using maximum likelihood estimation in IBM AMOS (Version 23.0; Arbuckle, 2014). Overall, the model fits the data well, χ²(df = 5) = 2.93, p = .71, χ²/df = 0.59, comparative fit index (CFI) = 1.00, root mean square error of approximation (RMSEA) = .00, standardized root mean square residual (SRMR) = .03. The path coefficients indicated the necessity of both theoretical perspectives for explaining disposition formation (see Figure 3).

OPEN IN VIEWER

Figure 3.

Path model results from Study 3.

Raney’s expansion

Consistent with Studies 1 and 2, the significant Path 1 indicates that the hero was perceived to be more moral (M = 3.78, SD = 1.25) than the villain (M = 3.10, SD = 0.96) absent any observable behavior, t(128) = 3.44, p < .001, Cohen’s d = 0.61.

Integral to Raney’s arguments are Path 2 and Path 4. Coupled together, these paths indicate a crossover interaction (see Figure 4). Although the 95% confidence intervals overlap, the significant interaction indicates that viewers may update their judgments based on observed behaviors and potentially overcorrect for initial impressions. Viewing the two characters engaging in similar behaviors led to relatively similar levels of approbation. However, the significant interaction suggests potential overcorrection for the villain: For the more moral behavior, the villain scored slightly higher (pairwise simple effect p = .08) on approbation than the hero.

OPEN IN VIEWER

Figure 4.

Visual depiction of crossover interaction effect of character-schema and behavior on moral approbation from Study 3.

The significant Path 5 is consistent with the Path 4 interaction and suggests that expectations shade subsequent perceptions of a character’s behaviors. The significant Path 6 indicates that the more moral a participant found the character to be prior to behavior, the more moral they found them afterward, suggesting that superficial judgments bias evidentiary judgments (i.e., those based on the observation of actual character behavior). Without this bias, this path should have been zero, indicating that moral judgments made prior to observing behavior were irrelevant for those made afterward. Thus, Raney’s (2004) expansion seems necessary for explaining disposition formation (see Online Supplement, Appendix E).

Finally, it is worth noting that the model provides evidence consistent with Sanders’s (2010) model of media character impression formation. Our empirical model suggests that character-schema led to initial impressions consistent with the schema. When characters were seen to engage in behavior, the activated schema interacted with the morality of the behavior to alter approbation processes. Next, approbation of behavior helped to refine initial impressions (seeing a moral action begets moral judgment, and vice versa), but the initial judgments were not completely overridden, despite the fact that they were based purely on visual impressions.

Character-Schema Activation From Visual Cues

That moral judgment of characters can precede the observation of behavior may seem inconsequential. After all, a person’s appearance can influence perceived trustworthiness (Buckingham et al., 2006), personality judgments (Shevlin, Walker, Davies, Banyard, & Lewis, 2003), and even the detection of social cheaters (Yamagishi, Tanida, Mashima, Shimoma, & Kanazawa, 2003). What makes our findings unique and noteworthy is the fact that visual cues associated with specific character-schemata were capable of eliciting judgments that precisely mirrored prototypes of these characters (see Eden et al., 2015). Had we simply observed that visual schema led to differences in moral judgment, warmth, duplicity, and competence, our findings would be attributable to differences in visual cues that may or may not have been associated with character-schema. However, because we observed differences in morality, warmth, and duplicity combined with a lack of difference in competence, we can be more confident that we activated character-schemata. That the presence of a rival character in Study 2 enhanced differences—without changing perceptions of competence—further increases our confidence. At the same time, there was an unexpected finding regarding the contrast effects.

In Study 2, our hero stimulus seemed to benefit from the presence of the opposing character more than the villain, as evidenced by the significant order by schema interactions. This asymmetrical benefit may relate to our success in creating the hero stimulus. When judged alone, our heroic character scored near the midpoint of the scale for all moral intuitions, indicating that although this character was not perceived to violate the moral intuitions, he also was not perceived to uphold them. Moreover, scores on “heroic” and “villainous” did not differ from each other for the heroic stimulus despite the fact that the hero was less villainous and more heroic than the villain. Thus, although Study 1 indicates that our hero was more moral, less villainous, and more heroic than our villain, it does not indicate that our hero upheld morality nor was he perceived to be a clear hero. These findings suggest that perhaps hero schema activation is more difficult than villain schema activation. Research from cognitive psychology indicates that negative information exerts stronger effects than positive information (see Baumeister, Bratslavsky, Finkenauer, & Vohs, 2001; Kahneman & Tversky, 1979). Thus, it may simply be easier to activate villain schema because humans have greater sensitivity to negative cues. At the same time, it may be the case that villains are more stable and less diverse in visual appearance as compared with heroes, leading to less ambiguous interpretations. This explanation is consistent with Campbell’s (1949/2004) influential work, The Hero With a Thousand Faces.

A central component of Campbell’s (1949/2004) thesis is the hero as “everyman/woman.” Campbell argues that heroes are indistinguishable from normal individuals at the outset of a tale, but through brave and virtuous actions, achieve heroic status. Thus, visual information alone may be less capable of activating hero schema as compared with villain schema because heroes are more visually varied than villains. Consider the varied appearance of three traditional heroes: Robin Hood, Luke Skywalker, and Frodo Baggins. Robin Hood is an adult, Luke Skywalker is a farm boy, and Frodo Baggins is a hobbit, perhaps the least visually heroic. Despite this variance, there is incredible consistency in their nemeses: All are clad in black robes and armor. Although anecdotal, this observation—coupled with Campbell’s analysis—seems to suggest differences in the discriminability of heroes and villains based on visual information.

The Effects of Character-Schema on Perceptions of Behavior

Raney’s (2004) moral justification approach and Zillmann’s (2000) moral judgment approach both seem necessary for explaining disposition formation. Social judgment theory’s (SJT; Sherif & Hovland, 1961) description of latitudes of acceptance may help explain when justification versus judgment is expected. Behaviors that deviate only slightly from schema-based anchors may simply be assimilated as consistent with the schema (i.e., moral justification). The larger the deviation, the more likely the behavior is to elicit moral judgment. However, schemas can still play a role here; they may lead to an overcorrection or contrast effect (see also expectancy violations theory; Burgoon, 2015). Writers are likely intuitively aware of what behaviors audiences might allow and those they would not (see Lee & Shapiro, 2014; Sanders, 2010). Still, social scientists could advance knowledge by probing the relationships between schema- and behavior-based moral judgments in experimental research.

Study 2’s contrast effects suggest that latitudes of acceptance might also be contingent upon comparisons made between characters (see Sanders, 2010). Consider, for example, the television series Breaking Bad. Throughout the series, the main character, Walter White, becomes increasingly evil, which left his creator, Vince Gilligan, surprised (TimesTalks, 2013):

As he got worse and worse . . ., I thought, “Well, hopefully the hook is set now and people keep watching even though they won’t like him as much . . .” And, lo and behold . . . 56 episodes in, people are still rooting for this guy . . . I would have guessed at this point he would have lacked sympathizability.

Gilligan’s surprise seems to ignore contrast effects and to assume that viewers evaluated Walter’s behaviors in relation to past behaviors or some concrete standard of morality. Perhaps viewers were making comparisons that continually allowed Walter’s behaviors to be cast in a more forgiving light. Throughout the series, the antagonists that Walter faced became increasingly dangerous and evil. His first antagonists were low-level drug dealers, and his behavior at that time—albeit morally questionable—was arguably more moral than theirs. By the series’ end, Walter became less moral compared with how he began and even compared with the initial antagonists. However, during this transition, Walter’s opponents also became more extreme; his final antagonists were literal Nazis. Thus, rather than being evaluated in a moral vacuum, Walter was likely being compared with his antagonists, allowing his behavior to meet minimal acceptable standards for viewers. In other words, Walter’s opponents may have moved Walter’s moral judgmental anchor. This process provides fertile ground for future work.

Finally, understanding these processes could hold value for entertainment-education campaigns. Characters are often used in entertainment-education contexts to serve as models for positive and negative focal behaviors (see Singhal & Rogers, 2002; see also Robinson & Knobloch-Westerwick, 2016). Our results regarding overcorrection of approbation for the villain suggest potential benefits of having an initially perceived “bad guy” model positive actions.

Prototypes, Exemplars, and Narrative Schema

Perhaps the most fruitful area of future research would involve incorporating cognitive psychology methods related to prototypes and exemplars (see Rosch, 1999) with narrative schema theories. Cognitive psychology research suggests that highly prototypical objects (e.g., a robin with respect to “birds”) are sorted into the category from which they originate more quickly than low prototypical objects (e.g., a penguin). The methods that uncovered this finding could be used to identify relationships between character-schemata. For example, Superman—a quintessential superhero—should be more quickly categorized as “hero” than Deadpool—an antihero. Various permutations of this procedure could help determine whether antihero characters (and other morally ambiguous characters [MACs]) are a subordinate category of the basic category “hero” or whether these characters fit into their own basic category distinct from both heroes and villains (see Krakowiak & Oliver, 2012; see also Krakowiak & Tsay-Vogel, 2015).

There is a growing focus within narrative research on MACs, such as antiheroes (see Eden, Daalmans, & Johnson, 2016; Eden, Grizzard, & Lewis, 2011; Janicke & Raney, 2015; Krakowiak & Oliver, 2012; Krakowiak & Tsay-Vogel, 2013, 2015; Raney & Janicke, 2013). A combination of our approach with this past work might yield novel predictions. For example, Lee and Shapiro (2014) provided a model suggesting a rationalist and intuitionist sequence for disposition formation. If MACs defy schema categorization, then a rationalist sequence may be necessary for disposition formation. However, an intuitive sequence may still be possible when a MAC is paired with an existing character-schema due to contrast effects.

Limitations

There are several limitations of the current project. As previously stated, our hero stimulus was somewhat ambiguous. We do not believe that this invalidates our findings for several reasons. First, although the “hero” character when judged alone was not perceived to be truly heroic, he was perceived to be more moral, more heroic, and less villainous than the villain. Second, the initial schema-based judgments in Study 3 were not entirely overridden by viewing actual behavior. At the same time, a cautious interpretation of the results of Study 3 is warranted. Future research should examine these processes using more robust hero schema.

A second concern is our stimuli’s relative lack of ecological validity. Many researchers argue that even a complete lack of ecological validity is not necessarily harmful for experimental work, which is designed to examine causal processes in controlled settings (see Mook, 1983). In addition, researchers who have used more ecologically valid stimuli (see Sanders & Tsay-Vogel, 2016) have called for carefully controlled experiments, such as those presented here. Still, it would be wise for future work to use narratives that feature both a hero and a villain—which may be more representative of Raney’s (2004) narrative schema logic.

Third, our studies have notable boundary conditions. Schema activation is more likely to occur in certain narratives (e.g., stand-alone films) as compared with others. For example, serial dramas and established characters are likely to elicit judgments derived not from visual schema but rather from previous exposures to the characters.

Finally, some might argue that our data simply reflect demand characteristics. However, if demand characteristics occurred, this is not a fatal flaw; schema activation is a form of demand characteristic. Second, demand is insufficient for explaining the totality of our data, particularly the discriminant validity associated with competence judgments of the characters. If our findings resulted purely from demand characteristics unassociated with schema activation, the villain should have also been perceived as being less competent, particularly in Study 2, which featured both characters.