Character Gender and Disposition Formation in Narratives: The Role of Competing Schema

ABSTRACT

How viewers form dispositions toward narrative characters is a central question of affective disposition theory. Two routes are explained by current models: Schema activation, whereby viewers’ dispositions are based on perceived narrative role, and behavioral approbation, whereby viewers’ dispositions are based on moral approval/disapproval of behavior. What remains unclear is how competing character schemas function: Do they exert their influence in the same location of the serial process? Or, does the impact of schemas on disposition formation depend on the schema? The current paper builds on past work that experimentally manipulated schema activation and behavioral approbation with experimental inductions. We extend that past work by crossing its hero/villain-schema induction with another: character gender. After validating stimuli in a pilot study, our main experiment demonstrated that gender did not moderate hero/villain-schema activation; behavioral approbation, however, was more extreme for female characters. Theoretical implications suggest that various character schemas may have distinct roles to play in disposition formation, with these distinctions being unaccounted for by current theory. Practical implications suggest that female characters may elicit stronger positive/negative dispositions and, through outcome evaluation processes, narrative enjoyment. Thus, Hollywood’s current lack of female character representation is likely hurting their bottom line.

Depictions of female characters in entertainment media have been a perennial topic of interest for researchers. In proportional terms, male narrative characters outnumber female characters roughly two to one (Annenberg Inclusion Initiative, 2020). When featured, female characters are depicted stereotypically (Beasley & Standley, 2002; Furnham & Mak, 1999; Jansz & Martis, 2007), are likely to be featured in secondary or supporting roles (Downs & Smith, 2010; Lynch et al., 2016), and reinforce negative attitudes toward women through detrimental portrayals (see Fox et al., 2015; Galdi et al., 2014). These systematic biases provide the groundwork for examining an untapped question for Zillmann’s (2000) affective disposition theory (ADT): To what extent do various character schemas operate similarly or differently in disposition formation processes?

Disposition formation relates to viewers forming perceptions of characters in moral terms. Behavioral approbation (i.e., approving or disapproving of the behavior of a character) was the primary route of disposition formation until Raney (2004) introduced a schema-activation route. Schemas are knowledge structures that facilitate recognition and categorization of environmental stimuli (see Ghosh & Gilboa, 2014, for a recent review). Through exposure to repeating patterns and the co-occurrence of features, individuals can build schemas related to any number of concepts. Raney argued that narrative consumers would develop narrative schemas for archetypal character roles (e.g., heroes/villains) and narrative structures (see Reagan, Mitchell, Kiley, Danforth, & Dodds, 2016). Although several studies have shown that viewers can spontaneously form dispositions toward schema-consistent characters (see Grizzard et al., 2020b; Grizzard et al., 2018), no research exists that examines how competing schemas operate during disposition formation. Because of the lack of research on competing schemas, ADT remains largely agnostic regarding whether different schemas might play different roles in disposition formation.

In the current paper, we examine these processes by pitting narrative-character schemas (i.e., hero/villain) against gender schemas (i.e., female/male) in studies built on Grizzard et al.’s (2018) disposition formation experimental paradigm (DFEP). Our studies seek to replicate findings from Grizzard et al. (2018; see also, Grizzard et al., 2020b) and determine (a) whether hero/villain-schema activation is distinct for female versus male characters and (b) whether character gender influences the behavioral approbation route of disposition formation.

The Affective Disposition Theory of Drama

ADT (Zillmann & Cantor, 1976; see also, Zillmann, 2000) is an important theory of narrative processing and evaluation that helps to explain viewers’ responses to narratives (see Raney, 2003, and Tamborini et al., 2021, for overviews). The theory describes how viewers come to view characters as moral/liked or immoral/disliked (i.e., disposition formation) and how their anticipated and realized fates elicit either euphoria or dysphoria (i.e., outcome evaluation). The focus of the current investigation is on disposition formation

According to ADT, disposition formation occurs through two routes: behavioral approbation and schema activation. Dispositions formed through behavioral approbation follow a serial causal process. First, viewers observe a character’s behavior. Then they approve or disapprove of the behavior. Finally, approbation converts into positive dispositions, and disapprobation converts into negative dispositions. Experimental tests are generally consistent with ADT’s propositions regarding behavioral approbation disposition formation. For example, Eden and Tamborini (2017) used a mundane story in which a character borrows a car, and they manipulated whether the car was returned with a full gas tank (i.e., a moral behavior) or with an empty gas tank (i.e., a less moral behavior). Mediation analyses were consistent with the aforementioned serial causal process. We thus propose our first hypotheses:

H1: More moral behavior will garner greater approbation than less moral behavior.

H2: Approbation of behavior will be positively related to perceived character morality.

The schema-activation route of disposition formation was posited by Raney (2004), who argued that disposition formation can occur through heuristic processes (see also, Krakowiak & Tsay-Vogel, 2015; Oliver et al., 2019). Through a history of narrative consumption, viewers develop understandings of stories that help them predict what will happen in the narrative and which characters are likely to be the heroes/villains. Writers often use narrative-character schema to aid in character development and meaning-making with the viewer. For example, villains are often clad in black clothing, and heroes in lighter colors (see Hoffner & Cantor, 1991). When visual cues reflect narrative-character schemas (e.g., “this character looks like a hero”), viewers can form dispositions without going through the behavioral-approbation route. Grizzard et al. (2018; see also, Grizzard et al., 2020b) have presented evidence of the schema-activation route in several studies whereby schema-consistent characters are presented to participants absent narrative context. When the characters are depicted in a manner consistent with heroic schema, participants judge them to be more moral than when the characters are depicted in a manner consistent with villainous schema.¹ We thus hypothesize:

H3: A heroic-looking character will be evaluated as more moral than a villainous-looking character.

Through multiple studies, Grizzard et al. (2018; see also, Grizzard et al., 2020b) have developed a Disposition Formation Experimental Paradigm (DFEP). The DFEP allows for an examination of both routes of disposition formation through temporally-dependent, experimental inductions and measurement strategies.

Grizzard et al.’s Disposition Formation Experimental Paradigm

Grizzard et al.’s DFEP begins by presenting participants with an image of a character whose visual depiction is manipulated to represent either a heroic or villainous schema. Past studies (see Grizzard et al., 2018; see also, Grizzard et al., 2020b) have manipulated four key visual features associated with heroes and villains (see Hoffner & Cantor, 1991): color of clothing (black for villain; lighter tones for hero), hair color (black for villain; brown or blonde for hero), facial expression (stern/angry for villain; happy/relaxed for hero), and facial scarring (present for villain; absent for hero). After viewing the character, participants rate their perceptions of the character’s morality, which allows for a test of the schema-activation hypothesis.

After initial ratings of the character are made, participants are presented with a short story that includes the previously-rated character. This simple narrative is manipulated such that the character behaves in either a more or less approved of manner. After reading the narrative and observing the character’s behavior, participants indicate their approval of the character’s behavior and rate their perceptions of the character’s morality again. These measurements allow for a test of the behavioral-approbation disposition formation hypotheses. Here we note that observing an indirect effect of the manipulated behavior on the general morality score is considered consistent with the serial causal process.

The combined manipulations and measures allow for disposition formation processes to be tested in a path model, which accounts for the effects of each route of disposition formation (see Figure 1). The path model includes a path (i.e., Path 1) to test the schema-activation hypothesis (i.e., H3). It also includes relevant paths (i.e., Paths 3 and 7) to test the behavioral-approbation hypotheses (i.e., H1 and H2). The remaining paths in the model account for other theoretical influences. Paths 2 and 4 test a potential additive and interactive influence of narrative-character schema on approbation. Path 5 tests a potential influence of preexisting moral judgments of the character on approval/disapproval of the character’s behavior in the narrative. Finally, Path 6 tests a potential influence of schema-based moral judgments (general morality T1) on final judgments of the character (general morality T2).

Figure 1. Path model from Grizzard et al. (2018) with participant sex added as a control variable

To effectively use this paradigm the following two conditions are required to be met. First, one must establish that the narrative-character schema manipulation induces variance in moral judgments of the character prior to the narrative (i.e., data are consistent with H3). Second, one must establish that the character behavior manipulation induces variance in behavioral approbation (i.e., data are consistent with H1). If these two conditions are met, one can be reasonably sure that they have established a situation in which schema activation occurs and there is adequate systematic co-variance between the randomly assigned behaviors and their approval/disapproval by participants.

Unanswered Questions Regarding Disposition Formation

Although Raney (2004) posited an influence of narrative-character schemas on disposition formation, ADT does not clearly delineate whether all schemas operate similarly in disposition formation. There is reason to believe that not all schemas are created equal with regard to disposition formation. Some schemas seem to have their origins in narratives and extend out to the real world, whereas others seem to have their origins in the real world and extend into narratives. For example, heroic and villainous schemas are central to narratives, but we often interpret real-world individuals or things in light of these schemas.² Other schemas are more central to the real world, but they still likely play a role in how we interpret narrative characters. For example, many demographic variables (e.g., age, race, gender) are readily observable in narrative characters, and schemas associated with these variables in the real world may alter how we perceive characters in narratives.

As it stands, ADT remains agnostic toward the potential of unique effects for different schemas. It simply assumes that schemas will have some effect on disposition formation processes. In the current studies, we examine whether the effects of different schemas influence disposition formation in unique ways. To do so, we adopt Grizzard et al.’s (2018) DFEP and layer on top of their experimental inductions an additional manipulation of the character’s gender. This extension allows us to examine (a) whether hero/villain-schema activation differs for male versus female characters and (b) whether behavioral approbation differs for male versus female characters. Past research on differential depictions of male and female gender roles in media entertainment provides logic for specifying potential moderating effects of character gender on schema activation and behavioral approbation.

Gender Roles and Their Potential Relation to Disposition Formation Processes

To explicate reasons why character gender might moderate disposition formation, we draw on research related to the prevalence of female characters as compared to male characters and punitive judgments associated with negative behaviors.

Prevalence and Portrayal of Female Characters and Schema Activation

Schema development relies on repeated exposure to similar patterns. Past research suggests that female characters are both underrepresented and relegated to secondary roles within narratives. Since 2007, Smith and colleagues (see Annenberg Inclusion Initiative, 2020, for the most recent report) have documented the prevalence of male and female speaking characters in the top 100 movies of the year. With a sample of 57,629 characters, the Annenberg Inclusion Initiative finds that male characters outnumber female characters at a rate of 2.2 to 1. Research on other media finds similar patterns of underrepresentation (see Aley & Hahn, 2020; Daalmans et al., 2017; Hamilton et al., 2006; Lynch et al., 2016; Signorielli et al., 1994). In addition, female characters, when present, are usually less central to the plot of the narrative. They typically are found in supporting and secondary roles rather than the most central protagonist/antagonist roles (which are often correlated with heroism and villainy; see Annenberg Inclusion Initiative, 2020; Hamilton et al., 2006; Lynch et al., 2016). Because of these facts, hero/villain-schema activation may be weaker for female characters as compared to male characters. This possibility is supported by research on children’s perceptions of heroes, which found that while male heroes had stereotypically male traits, female heroes did not have stereotypically female traits (see Holub et al., 2008). At the same time, there is reason to expect that hero/villain-schema activation may not be moderated by character gender. Repeated exposure to specific female heroes/villains could result in equally strong hero/villain-schema activation for female characters despite their underrepresentation. For example, repeated exposure to popular narratives with strong female heroes and villains (e.g., She-Ra, Wonder Woman, Wizard of Oz) could lead to equally strong heroic/villainous-schema formation for female characters. Thus, we question:

RQ1: Will hero/villain-schema activation be weaker for female characters as compared to male characters?

Gender Biases and Behavioral Approbation

Punishment for moral transgressions is particularly relevant to narrative evaluations (see Bilandzic et al., 2017; Daalmans et al., 2017; Grizzard et al., 2021a; Raney & Bryant, 2002; Rothmund et al., 2013). Thus, we draw on criminal justice research on gender biases related to disapproval processes in the real world to form a hypothesis regarding how character gender might moderate the behavioral approbation process. Women in the criminal justice system receive significantly shorter sentences than men (see Starr, 2015; U.S. Sentencing Commission, 2017). Women are also more likely to avoid charges, conviction, and incarceration (see Starr, 2015), and these findings hold up under meta-analysis (see Bontrager et al., 2013). Given that sentences for criminal acts are a form of disapprobation for immoral (or at least unlawful) behavior, the patterns in the literature suggest that male characters – if evaluated similarly to real-world men – would receive greater disapprobation for immoral actions as compared to female characters. We thus hypothesize:

H4: The effect of moral/immoral behavior on behavioral approbation will be weaker for female characters as compared to male characters.

The Current Studies

Pilot Study

Grizzard et al.’s (2018) DFEP has already been validated with male characters (see also Grizzard et al., 2020b), and we used male characters from Grizzard et al. (2020b) as the male characters in the current study. To ensure that female characters were indeed capable of activating hero/villain schemas, we conducted a pilot study intending to examine schema-activation for two sets of heroic and villainous male and female characters. Results of the pilot study replicated the hero/villain-schema activation findings from Grizzard et al. (2020b) and also demonstrated that narrative-character schema activation can occur for both male and female characters (see the online supplement for complete method and results; http://doi.org/10.17605/OSF.IO/NGQY8). Based on these results, we moved ahead with our main study.

Main Study Design and Materials

Replicating and extending Grizzard et al. (2018, see Study 3), our main study was a 2 (narrative-character schema: heroic- versus villainous-looking character; between-subjects) X 2 (character behavior: more moral versus less moral; between-subjects) X 2 (character gender: male versus female; between-subjects) design.

Stimuli

Our validated character stimuli are presented in Figure 2. To manipulate character gender, we began by creating the same male heroic and villainous characters used in Grizzard et al. (2020b), which were shown to activate strong schematic-perceptions of heroes and villains. We then created female versions of each using the same software and changing the character’s gender from male to female. This change resulted in a lengthening of the hair of the character and a feminizing of the facial features (i.e., a less angular chin) and torso (the addition of female breasts and reduction in the breadth of the shoulders).

Figure 2. Male and female hero and villain stimuli for the main study

To induce variance in behavioral approbation and allow direct comparisons with past research, we used the same narrative as Grizzard et al. (2018). The narrative presents a version of the Trolley Problem (see Foot, 1967). The character finds themselves in a situation in which an out-of-control trolley is heading toward five workers threatening their lives. Importantly, there is a switch-track lever next to the character that can send the trolley down a sidetrack where only one person is working. The character decides (based on random assignment) either (a) to pull the lever, saving the five workers but killing the one or (b) to do nothing, resulting in the five workers’ deaths. Although the narrative is a moral dilemma, research indicates that most individuals view pulling the lever to be a more moral act than doing nothing (see Greene et al., 2009; Grizzard et al., 2018, 2020b; Thomson, 1985). Thus, even though pulling the lever may not be objectively moral and not pulling the lever may not be objectively immoral, the effect size associated with behavioral approbation of these two behaviors in past usage of this narrative is quite large (Cohen’s d = 1.23 in Grizzard et al., 2018; Cohen’s d = 1.55 in, Grizzard et al., 2020b). This large effect size indicates that one behavior is approved of far more than the other. Thus, this narrative and its randomly assigned behavior for the character is likely to satisfy the aforementioned conditions for testing the behavioral approbation hypotheses. The full stimuli are presented in our online supplement (http://doi.org/10.17605/OSF.IO/NGQY8).

Measures

Character morality was assessed following initial presentation of the character (Time 1; T1) and following the narrative (Time 2; T2) with the Extended Character Morality Questionnaire (CMFQ-X; Grizzard et al., 2020a). Based in moral foundations theory (MFT; Haidt & Joseph, 2007), the CMFQ-X consists of the prompt “The character seems like they would … ” followed by specific immoral (e.g., “cause someone to suffer emotionally”) or moral (e.g., “treat people equally”) actions, each associated with a specific moral foundation.³ Responses to each action are rated on a 7-point Likert-type scale. Reliabilities for the subscales are as follows: care (α_T1 = .91, α_T2 = .89), fairness (α_T1 = .84, α_T2 = .84), loyalty (α_T1 = .77, α_T2 = .81), authority (α_T1 = .78, α_T2 = .82), and purity (α_T1 = .68, α_T2 = .71). Consistent with past work (see Grizzard et al., 2018, 2020b), we created a general morality composite (α_T1 = .91, α_T2 = .94) by averaging across the five subscales. To measure approbation of behavior, we used the same measure from Grizzard et al. (2018). Participants rated their agreement on a 7-point Likert-type scale for nine statements. Example statements include “[The character’s] behavior was … ” “ethical,” and “wrong” (reverse-coded). Reliability for the measure was high (α = .91), and we averaged across items to create an approbation of behavior composite. Results regarding the measurement model tests of the scales are presented in the online supplement.

Procedure

Participants first granted informed consent. Next, they viewed their randomly assigned character and completed the general morality measure (T1). Next, they read the narrative with the randomly assigned behavior. Finally, they completed the approbation of behavior measure and the general morality measure (T2) again.

Participants

Participants (N = 392) were recruited from a large, public university in a northeastern state in the U.S. Participants received course credit for participation, and all procedures were determined exempt by the local Institutional Review Board. Data collection occurred online from February 14 to March 19, 2018. Several steps were conducted to ensure data quality. First, participants who entered the survey but failed to complete 90% of the survey were dropped (n = 23). Next, several attention checks were included. These were the name of the character, the subject of the story, and the behavior the character enacted. Participants (n = 99) who failed to complete all three attention checks accurately were dropped (final N = 270; n_male = 128, 47.4%; n_female = 142, 52.6%; M_Age = 20.24, SD_Age = 2.29.). The survey software randomly assigned participants to one of our eight conditions. Random assignment resulted in roughly equivalent conditions in terms of cell sizes: male hero, more moral behavior = 30; male hero, less moral behavior = 24; male villain, more moral behavior = 36; male villain, less moral behavior = 33; female hero, more moral behavior = 36; female hero, less moral behavior = 36; female villain, more moral behavior = 38; female villain, less moral behavior = 37.

We conducted power analyses for our manipulations using G*Power. We used effects sizes reported by Grizzard et al. (2018) for character schema (Cohen’s d = .61) and character behavior (Cohen’s d = 1.23) as the bases of our power analyses. We utilized the t-test family in G*Power given that our manipulations were simple between-subjects manipulations and t-tests are the basis for testing the significance of regression coefficients in a path model. For character schema, the power analysis indicated a sample size of 88 was necessary with power (1-β) set to .80 and alpha error set to .05 (two-tailed). For character behavior, the power analysis indicated a sample size of 24 with power (1-β) set to .80 and alpha error set to .05 (two-tailed).

Results

Manipulation Checks

We conducted manipulation checks on our narrative-character schema and our character behavior inductions by conducting a multivariate analysis of variance (MANOVA) with the independent factors being our three experimental inductions (character gender, narrative-character schema, and character behavior) and the dependent variables being (a) the general morality T1 score (in order to assess the strength of the narrative-character schema induction) and (b) the approbation of behavior score (in order to assess the strength of the character behavior induction). Box’s Test of Equality of Covariance Matrices was significant, Box’s M = 34.51, F(21, 195831.76) = 1.60, p = .04, and so Pillai’s Trace was interpreted. Results were not affected. Multivariate results are presented in Table 1. The only factors that had significant effects on the dependent variables were the narrative-character schema factor and the character behavior factor. All other factors and interactions were nonsignificant.

Table 1. Multivariate results from the MANOVA examining the effects of the experimental inductions on general morality T1 and approbation of behavior

Factor	Pillai’s Trace	F(2,261)	p	η^2p
Ch. Gender	.01	1.19	.31	.01
Narrative Ch. Schema	.38	78.54	<.001	.38
Ch. Behavior	.25	42.27	<.001	.25
Ch. Gender X Narrative Ch. Schema	.02	1.93	.15	.02
Ch. Gender X Ch. Behavior	.01	1.80	.17	.01
Narrative Ch. Schema X Ch. Behavior	.01	1.27	.28	.01
Narrative Ch. Schema X Ch. Behavior X Ch. Gender	.00	0.53	.59	.00

The between-subjects effects are presented in Table 2. Notably, Levene’s test of equality of error variances was nonsignificant for both general morality T1 (p = .13) and approbation of behavior (p = .08). The between-subjects results indicated that our inductions worked as intended and in an isolated manner. The narrative-character schema induction had a significant effect on general morality T1 and only on general morality T1. Consistent with expectations, the heroic-looking character was perceived to be more moral (estimated marginal M = 4.48, SE = 0.08) than the villainous-looking character (estimated marginal M = 3.20, SE = 0.07). The character behavior induction had a significant effect on approbation of behavior and only on approbation of behavior. Consistent with expectations, the estimated marginal means indicated that pulling the lever was approved of more (M = 4.99, SE = 0.09) than not pulling the lever (M = 3.85, SE = 0.09). Effect sizes for the effects of these inductions were large (η² =.37, η²_p =.38; and η² = .24, η²_p = .24, respectively).

Table 2. Univariate results examining the effects of the experimental inductions on general morality T1 and approbation of behavior

Dependent Variable: General Morality T1
Factor	F(1,262)	p	η^2	η^2p
Ch. Gender	2.21	.14	.01	.01
Narrative Ch. Schema	157.63	<.001	.37	.38
Ch. Behavior	0.73	.40	.00	.00
Ch. Gender X Narrative Ch. Schema	3.02	.08	.01	.01
Ch. Gender X Ch. Behavior	0.14	.71	.00	.00
Narrative Ch. Schema X Ch. Behavior	2.18	.14	.01	.01
Narrative Ch. Schema X Ch. Behavior X Ch. Gender	0.26	.61	.00	.01
Dependent Variable: Approbation of Behavior
Factor	F(1,262)	p	η^2	η^2p
Ch. Gender	0.17	.68	.00	.00
Narrative Ch. Schema	0.07	.79	.00	.00
Ch. Behavior	84.09	<.001	.24	.24
Ch. Gender X Narrative Ch. Schema	0.84	.36	.00	.00
Ch. Gender X Ch. Behavior	3.47	.06	.01	.01
Narrative Ch. Schema X Ch. Behavior	0.37	.54	.00	.00
Narrative Ch. Schema X Ch. Behavior X Ch. Gender	0.81	.37	.00	.00

Main Analyses

To test our hypotheses and answer our research question, we conducted two path analyses. Both analyses controlled for participant sex; all findings, however, were robust to the inclusion or exclusion of participant sex as a control variable. The first path analysis (Model 1) consisted of testing the model in Figure 1 for all participants regardless of character gender. Following this analysis, we conducted a multiple groups analysis that tested the model in Figure 1 for each character gender separately. The first path analysis provides a general test of schema activation and behavioral approbation disposition formation processes by replicating Grizzard et al.’s (2018) analysis and assessing both the significance of the relevant paths in the model and the overall fit of the model to the data. The second, multiple groups path analysis provides an examination of whether character gender moderates the disposition formation processes. Moderation is assessed by examining two indicators. First, we compare whether a model which holds the paths equal between the groups fits worse than an unconstrained model. Finding such a decrement in fit indicates that the processes represented in the model differ for male characters versus female characters. Second, we examine whether each path in the model is significantly different between the two groups. Finding differences between the groups in specific paths localizes the moderating impact of character gender. Path analyses were conducted in AMOS and used maximum likelihood estimation. Notably, our data did not violate the multivariate normality assumption. Still, assessment of the paths and the indirect effects were based on 10,000 bootstrap samples with 95% bias-corrected confidence intervals to account for any potential violations and resulting increased Type-I error rates related to deviations from this assumption.

Model 1 – Replicating Grizzard et al. (2018)

Results of the Model 1 path analysis indicated that the model fit the data well (see Figure 3). The significant path from the behavior manipulation to approbation of behavior (.53), coupled with the significant path from approbation of behavior to general morality T2 (.57) and the significant indirect effect from the behavior manipulation to general morality T2 (.31) indicate consistency with H1 and H2. These results replicate the manipulation check regarding character behavior and indicate that dispositions are formed along the behavioral approbation route of ADT. Overall, these results are consistent with Grizzard et al. (2018).

Figure 3. Depiction of the path analysis results. Beta weights for the paths are standardized estimates

The significant path from narrative-character schema to general morality T1 (.60) replicates the narrative-character schema manipulation check and is consistent with H3. This result indicates that dispositions are formed along the schema-activation route of ADT. This result is also consistent with Grizzard et al. (2018).

Notably, there was one significant path from participant sex to an endogenous variable, general morality T1. This path indicates that male participants ranked the character (regardless of condition) as more moral than female participants. Although this finding justifies the use of participant sex as a control variable, we note that the relationship here is far weaker than the paths related to disposition formation.

Only two paths from the current study failed to replicate those of Grizzard et al. (2018). The first was the path from the Narrative-character Schema X Behavior interaction to approbation of behavior, which was nonsignificant in the current study, but significant in Grizzard et al. (2018). The other was the path from general morality T1 to approbation of behavior, which was also nonsignificant in the current study, but significant in Grizzard et al. (2018). That said, the study by Grizzard et al. (2020b) used the same narrative and a similar analysis also found these paths to be nonsignificant. Thus, these results are consistent with that past work. Overall, two of the three studies utilizing these stimuli have found nonsignificant paths here, which indicates either a weak effect that requires a larger sample size for significance, or the results of Grizzard et al. (2018) for this path were a Type I error.

In general, the first model test indicates that the data are consistent with both forms of disposition formation. We find that heroic-looking characters are rated to be more moral than villainous-looking characters, explaining 38% of the variance in initial perceptions of character morality. In addition, the more moral decision from our stimulus was rated as such, with the manipulation explaining 25% of the variance in approbation of behavior. Finally, the model was capable of explaining 38% of the variance in final moral perceptions of the character.

Model 2 – Assessing whether Character Gender Influences Disposition Formation

Overall, the unconstrained multiple groups model fit the data well (see Figure 4). We found that the version of the model which held structural weights equal resulted in a worse fit than the unconstrained model, Δχ² (Δdf = 10) = 25.03, p = .01. Given this result, we compared the path coefficients for male versus female characters to examine how and where character gender moderates the disposition formation processes.

Figure 4. Depiction of the multiple group analysis results. Beta weights for the paths are standardized estimates

Three path coefficients were significantly different between character genders (see Figure 4): (1) the path from the behavior manipulation to approbation of behavior, (2) the path from general morality T1 to general morality T2, and (3) the path from general morality T1 to approbation of behavior. That said, the path from general morality T1 to approbation of behavior was nonsignificant for both male and female characters, and so we view this moderation as unimportant for understanding how disposition formation differs between male and female characters.⁴

The path from the behavior manipulation to the approbation of behavior measure was significantly stronger (Z = 2.23, p = .03) for female characters (.65, SE = .09) as compared to male characters (.39, SE = .12). This significant difference indicates that the difference in approval between the more and less moral behavior was significantly larger for female characters as compared to male characters.

The path from general morality T1 to general morality T2 was also significantly stronger (Z = 2.33, p = .02) for female characters (.33, SE = .07) as compared to male characters (.14, SE = .09). Conceptually, this path indicates whether participants updated their beliefs about characters in a systematic or unsystematic manner. A positive significant path indicates that participants maintained their judgment across time, a negative significant path would indicate that participants reversed their judgment across time, and a zero path would indicate that updating was unsystematic. The positive significant path for female characters, the nonsignificant path for male characters, and the significant difference between the two suggests that participants were (a) more likely to maintain their moral judgments of female characters and (b) more willing to completely update their moral judgments of male characters. This difference in updating may indicate there is a greater influence of narrative-character schema-based dispositions on final dispositions for female characters as compared to male characters. Such an interpretation would suggest that perceptions of female characters are more predictable and stable than those of male characters.

Examining the Equivalence of the Nonsignificant Paths Related to Disposition Formation

Our previous analyses indicate that there are significant differences between male and female characters in the approbation of moral/immoral behavior, and nonsignificant differences between male and female characters with regard to (1) narrative-character schema activation (Z = −1.80, p = .07) and (2) the path from behavioral approbation to general morality T2 (Z = 0.23, p = .82). Of course, nonsignificant differences do not indicate equivalence, and so we conducted equivalence tests on these paths to determine whether they are statistically equivalent. We used Lakens et al.’s(2018) two one-sided tests procedure (TOST) for these tests.

For the required effect size estimate for the equivalence test on the narrative-character schema manipulation to general morality T1, we calculated the effect size difference between male and female characters for the path from our behavior manipulation to approbation of behavior (Cohen’s d = |.35|), which provides a point estimate for the effect size related to a significant difference between characters. This effect size is slightly larger than a moderate effect. Results of the equivalence test were nonsignificant, t(258.04) = −1.38, p = .08, and the null hypothesis test result was nonsignificant, t(258.04) = 1.53, p = .13. Based on these results, we have evidence to conclude that the difference in the path from our narrative-character schema manipulation to general morality T1 is statistically not different from zero; however, we lack evidence to conclude that the difference is equivalent to zero.

We repeated this equivalence test procedure to determine whether the path from approbation of behavior to general morality T2 was equivalent for male and female characters, and we used the same estimate of effect size. Results of the equivalence test here were significant, t(250.33) = 2.39, p = .009, and the null hypothesis test result was nonsignificant, t(250.33) = −0.47, p = .642. Based on these results, we have evidence to conclude that the difference in the path from our approbation of behavior measure to the general morality T2 measure is statistically not different from zero and statistically equivalent to zero.

Summary of Results

Our analyses find evidence that (a) any differences in narrative-character schema activation between male and female characters are minute, but not necessarily equivalent, (b) there is a significant difference in the first half of the behavioral approbation route for male characters versus female characters, and (c) the second half of the behavioral approbation route is equivalent for male characters and female characters. These findings indicate that not all schemas function similarly with regard to disposition formation processes. Narrative-character schemas related to heroism and villainy exerted their influence in the path from the narrative-character schema manipulation to general morality T1. Character gender schemas related to male and female characters exerted a moderating influence in the path from the character behavior manipulation to approbation of behavior, but not in the path from approbation of behavior to general morality T2.

General Discussion

The current study examines how different character schema might interact with and influence disposition formation processes. Our findings indicate that heroic/villainous-schema activation occurs similarly for both male and female characters. Heroic and villainous visual features seem capable of activating moral and immoral perceptions of a character for both male characters and female characters. At the same time, our findings indicate that disposition formation through the behavioral-approbation route differs for male characters versus female characters. The effect of character behavior on approbation was significantly stronger for female characters as compared to male characters, seeming to indicate that when the same behaviors are enacted by male and female characters, female characters garner greater approval/disapproval.

Theoretical Implications Related to ADT

Our findings indicate that ADT’s descriptions of disposition formation are slightly underdeveloped. Differential effects of specific schema are not considered or encapsulated in current disposition formation models. Although some theoretical models explain how schema are abandoned when observed behaviors deviate from expectations (see Sanders, 2010), ADT does not specify where or how specific schema might influence disposition formation. A systematic accounting of how various character schema influence disposition formation could aid in theoretical development by improving the specification of theoretical predictions made by ADT (see DeAndrea & Holbert, 2017) as well as the statistical models that explicate those predictions. As theories advance, the fidelity of mathematical and statistical models that operationalize conceptual processes becomes increasingly important for scientific research (see Huskey et al., 2020; Marr, 1982; van Rooij & Baggio, 2020).

Our findings suggest that hero/villain schema influence initial judgments of a character’s morality whereas gender-based schema influence the approbation of character behavior. Do our findings reflect a generalized process wherein character demographic variables (such as character gender) influence behavioral approbation and narrative-relevant variables (such as heroic/villainous visual efdepiction) influence schema activation? We think it is too early for such a conclusion. For example, some demographic variables, such as a character’s age, might influence schema activation through heroic/villainous visual depiction. Although audiences are just as likely to accept a female character as a hero/villain as a male character, they might be less willing to accept that a young child or fidelderly person could be either a hero or a villain as compared to a middle-aged adult. Thus, future research is needed to develop a fuller understanding and accounting of ADT’s disposition formation processes related to character schema. Using similar procedures as the current study (e.g., Grizzard et al.’s [2018] DFEP) and crossing various character schema through experimental manipulations would aid in this endeavor.

Theoretical Implications for Research on Bias

The current research design may also have some implications for studies and theories focused on gender-based biases. Our findings with regard to behavioral approbation did not align with most biases associated with gender. Rather than receiving relatively less approbation/disapprobation as compared to male characters, female characters received more. This reversal of the typical gender bias has been observed in some other contexts. For example, Johnson et al., (2018) found that in crowd-funding contexts (both in the lab and in naturalistic settings) women entrepreneurs were more likely to be funded than men. Rather than being based in ADT, Johnson et al. (2018) drew strongly on the stereotype content model (Fiske et al, 2002), which ignores morality judgments, and focuses on competence and warmth judgments. Grizzard et al.’s (2018) DFEP could provide a framework for understanding how predictions from complementary theoretical approaches – such as, the stereotype content model (Fiske et al., 2002) and other person-perception models (for review, see Moskowitz & Gill, 2013) – play out in mediated settings. Mediated settings can be a particularly useful for examining cognitive processes as “they are deliberately crafted products meant to elicit particular human thought, emotion, and behavior” (see Grall & Finn, 2021).

We also posit a use of the current experimental paradigm that might relate to implicitly measuring demographically-related bias. A within-subjects study whereby participants evaluated a series of heroic- and villainous-looking stimuli with varying demographic characteristics (e.g., race, gender-identity) could be used to generate within-subject effects associated with the activation of hero/villain schema. If the effect sizes for heroic/villainous schema are smaller for some demographic characteristics (e.g., Black, female characters) as compared to others (e.g., white, male characters), this difference might indicate a bias in the mind of the participant. Of course, this is a controversial and speculative suggestion, but empirical research could explore it in more detail.⁵

Practical Implications

The current research also has some notable practical implications related to Hollywood casting decisions. As stated in the introduction, the Annenberg Inclusion Initiative (2020) has documented a large gender gap in Hollywood casting decisions. Men are cast far more often in leading roles than women. The lead researcher of the project Stacy Smith argues that these disparities might relate to a fear in the movie industry that female actors are likely to be less of a draw than male actors (see TED, 2017). Our findings suggest that these fears may not only be exaggerated, but entirely unfounded. An important process related to the enjoyment of media narratives is the development of strong dispositions (see Raney, 2003). Seeing a hated villain thwarted at the hands of a beloved hero will elicit greater enjoyment than seeing a moderately disliked villain thwarted at the hands of a moderately likeable hero. Our findings suggest that narrative-character schema activation (i.e., heuristic judgments of a character as a hero or villain) was not systematically different for female characters as compared to male characters. Moreover, the first link in the behavioral approbation disposition formation process was more extreme for female characters as compared to male characters. Thus, we might expect that it is easier for writers to create highly moral, likeable heroes and highly immoral, dislikable villains when the characters are female as compared to male. In total, these results indicate that the underrepresentation of women in narratives could result in unrealized financial gains.

Limitations and Future Research Directions

Several limitations and future research directions warrant discussion. One element that we did not explore in our current study is whether character gender might have systematic effects on the perception of a character’s centrality to the plot, which we alluded to earlier in the paper. Because our study designs utilized only one character and character gender was always manipulated between-subjects, they were not well-equipped to explore whether perceived centrality is affected by character gender. Future research should utilize a larger cast of characters with measures specifically designed to assess perceived character centrality.

Our work and the past work of Grizzard et al. (2018; see Grizzard et al., 2020b) has focused on schema activation through visual means. Future work would be needed to determine if other narrative features are capable of activating schemas and whether these other schema-activating features have differential effects between male and female characters.

A third limitation relates to the outcome evaluation processes specified by ADT. In addition to disposition formation, ADT posits an outcome evaluation process whereby the outcomes that befall characters facilitate enjoyment. We find that character gender influences disposition formation in specific ways. How character gender influences outcome evaluation requires additional studies that manipulate both dispositions and the outcomes (i.e., the rewards and punishments) that befall characters.

A final limitation relates to our use of the Trolley Problem. There is a growing need for other stimuli to examine behavioral approbation disposition formation processes as the Trolley Problem becomes more well known. Of course, the ramifications of our use of the Trolley Problem is not a major issue as we instilled the intended systematic variance in behavior approbation. That said, future research should seek out new stimuli that cleanly manipulate the behavioral approbation route of disposition formation in order to improve the generalizability of our findings and to ensure that familiarity with the Trolley Problem did not hamper our ability to observe other processes at play.

We close this section by pointing to other important theoretical advances made to ADT in recent years that have particular relevance for our findings. First, Grizzard et al.’s (2020b) interdependence model of disposition formation describes how features of separate characters interact and influence each other. Dispositions formed toward one character are influenced not only by the features or behaviors of that character, but also by the features and behaviors of other characters within the narrative (see also, Grizzard et al., 2021b). Future work should thus consider how the pairing of male and female characters together influence both routes of disposition formation. A second theoretical advance of ADT with particular relevance to the current study is Tamborini et al.’s (2018) integration of attribution theory. Tamborini et al. (2018) argue that the attributions viewers make regarding character behavior influence approbation. When viewers attribute a character’s behavior to the character’s own desires (i.e., an internal attribution), viewers are likely to harshly judge any immoral action of the character. However, when viewers attribute a character’s behavior to some external force (i.e., an external attribution), viewers become willing to overlook the immoral actions of the character that occur due to this external force. Our findings related to character gender and behavioral approbation may indicate differences in attributions that participants made for the character’s behavior. Given the importance of Tamborini et al.’s (2018) attribution theory framework for identifying boundary conditions for some of ADT’s predictions, future research would be wise to consider how attributions of character behaviors differ based on the demographics of the character.

Conclusion

The current paper examined how character gender influences disposition formation. We found that female characters, like male characters, are capable of eliciting heroic and villainous perceptions, and schema activation for female characters is not systematically reduced as compared to male characters. We also found that participants were more sensitive to female-character behaviors than male-character behaviors. This resulted in more extreme approbation of behavior. This manuscript suggests a generative line of research related to exploring the role of different schemas in disposition formation. What role do various schema play in the disposition formation process, and can explicating these roles help to improve the explanatory power of ADT? By building on our study designs, future research can begin to specify and generalize the processes at play when competing schema are present in characters.

Open Scholarship

This article has earned the Center for Open Science badge for Open Data. The data are openly accessible at https://doi.org/10.17605/OSF.IO/NGQY8.

Acknowledgments

The authors would like to acknowledge Dr. Teresa Lynch for her helpful comments on an earlier draft of this article. The authors would also like to acknowledge the three anonymous reviewers and the associate editor, Dr. Benjamin Johnson, for their vital feedback during the review process.

Disclosure Statement

No potential conflict of interest was reported by the author(s).

Data Availability Statement

The data described in this article are openly available in the Open Science Framework at https://doi.org/10.17605/OSF.IO/NGQY8.

Notes

1. See the online supplement (http://doi.org/10.17605/OSF.IO/NGQY8) for more information about this operationalization of schema-activation.

2. The extension of heroic and villainous schemas to real-world phenomena is even present in the academic literature. For example, Gieniec et al., (2019) composed a scientific article that questioned whether cancer-associated fibroblasts were heroes or villains, and Morse (2011) composed an economics article that questioned whether payday lenders were heroes or villains. Finally, Hanke et al. (2015) explored how historical figures – such as Albert Einstein, Mother Teresa, Adolf Hitler, and Osama bin Laden – are conceptualized to be heroes or villains.

3. One of the purity items for the full scale (“be a smoker”) was inadvertently omitted during data collection. The reliability of the scale did not seem to be affected by this omission and is similar to those of the scale’s validation paper (see Grizzard et al., 2020a). In addition, measurement model tests (see online supplement) indicate that the unidimensional measurement models of the scales are consistent with the measurement models in the validation paper.

4. The path from general morality T1 to approbation of behavior was significantly more positive (Z = −2.76, p = .01) for male characters (.23) as compared to female characters (−.14). That said, this path was nonsignificant for both groups (see Figure 4). This may indicate that the influence of general morality T1 on approbation of behavior is a trivial effect or it could be the case that the current findings are the result of a Type II error and a lack of power. Future research is needed to disentangle these competing explanations, but we do note that the study was sufficiently powered to detect the significance of other paths in the model.

5. Matthews (2019) has developed a method for detecting boundaries related to disposition formation biases related to narrative-character schema. This method might also be useful for examining disposition formation baises related to character demographics.