FIGHTING ZOMBIES TOGETHER: A LONGITUDINAL EXPERIMENTAL TEST OF PREJUDICE REDUCTION THROUGH MEDIA EXPOSURE TO NON-HUMAN VILLAINS By Xuejing Yao A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of Communication – Doctor of Philosophy 2022 ABSTRACT FIGHTING ZOMBIES TOGETHER: A LONGITUDINAL EXPERIMENTAL TEST OF PREJUDICE REDUCTION THROUGH MEDIA EXPOSURE TO NON-HUMAN VILLAINS By Xuejing Yao Increasing the salience of shared human identity has been demonstrated as an effective mechanism to reduce interracial prejudice (e.g., Ellithorpe et al. 2018; Yao et al., 2022a). Previous research has found that watching racially diverse human heroes fighting against non- human villains increased the strength of viewers’ human identity, and their stronger human identity is in turn associated with more positive attitudes toward racial minority groups. This dissertation replicated Ellithorpe et al.’s (2018) research through a longitudinal controlled experiment with three waves of stimuli exposure and a posttest. The longitudinal design also made it possible to test the chronic accessibility of the human identity. Overall, minimal support was found for the longitudinal mediation with the current data and the human identity showed relatively short salience. However, this set of findings provided important information about the property of the superordinate human identity. Implications of the current findings as well as directions for future research were discussed. Keywords: common ingroup identity, prejudice reduction, media effects, supernatural genre This dissertation is dedicated to Mom and Dad. Thank you for believing in me. 此文献给爸爸妈妈。 iii ACKNOWLEDGMENTS Thank you to Dr. Dave Ewoldsen, who has trusted me as a fellow researcher long before I trusted myself. Thank you to Dr. Nancy Rhodes, who edited my research papers so many times but still patiently said “the paper is getting close” at every round. Thank you to Dr. Morgan Ellithorpe, who has shown me it is possible to have a successful career and a loving family at the same time. Thank you to Dr. Steve Pierce and my other colleagues at CSTAT, whose knowledge on statistics has frequently reminded me how little I know and motivated me to keep learning. Thank you to my friends, whose support and encouragement has constantly reminded me that I am not alone in this long academic journey. To name a few of such friends (and names are not in particular order): Kelsey, Sun, Youjin, Nikki, Sam, Shelby, Siyuan, Joomi, Lillie, and Alex. Thank you to Reed, who spent hours proofreading my dissertation, who has also spent years by my side and has been my rock. Thank you to my parents, who always go out of their way to celebrate every success of mine, no matter how trivial the success is. iv TABLE OF CONTENTS LIST OF TABLES ......................................................................................................................... vi LIST OF FIGURES ...................................................................................................................... vii KEY TO ABBREVIATIONS ...................................................................................................... viii CHAPTER 1 ................................................................................................................................... 1 INTRODUCTION ...................................................................................................................... 1 Self-Categorization and Ingroup Favoritism .......................................................................... 3 Common Ingroup Identity Model ....................................................................................... 4 The Present Study ................................................................................................................... 5 CHAPTER 2 ................................................................................................................................. 11 METHOD ................................................................................................................................. 11 Participants & Procedure ...................................................................................................... 11 Design ................................................................................................................................... 14 Experimental Stimuli ........................................................................................................ 14 Measures ........................................................................................................................... 17 CHAPTER 3 ................................................................................................................................. 21 STATISTICAL ANALYSIS .................................................................................................... 21 CHAPTER 4 ................................................................................................................................. 25 RESULTS ................................................................................................................................. 25 CHAPTER 5 ................................................................................................................................. 28 DISCUSSION ........................................................................................................................... 28 Limitations and Future Research .......................................................................................... 33 Conclusion ............................................................................................................................ 35 REFERENCES ............................................................................................................................. 37 v LIST OF TABLES Table 1 Overview of the Longitudinal Experiment Design……………………………….. 12 Table 2 Descriptive Statistics……………………………………………………………… 17 Table 3 Hypothesized Direct and Indirect Effects………………………………………… 26 Table 4 Mediation Effects within Each Wave…………………………………………… 31 Table 5 Post Hoc Analyses for RQ 1-3 with Data from the Posttest (W4)………………... 33 vi LIST OF FIGURES Figure 1 Path Model……………………………………………………………………….. 22 vii KEY TO ABBREVIATIONS CIIM Common Ingroup Identity Model CII Common Ingroup Identity SCT Self-Categorization Theory SIT Social Identity Theory SEM Structural Equation Modeling CLPM Cross-Lagged Panel Model FIML Full Information Maximum Likelihood viii CHAPTER 1 INTRODUCTION From Sleepy Hollow to The Walking Dead, the supernatural genre on television has been very popular in the past decade. For example, The Walking Dead, which features zombies as the major threat to the main characters, attracted more than 10 million viewers for a majority of its 11 season premieres (e.g., Thumbore, 2017). There are other examples of supernatural and science fiction series with non-human characters drawing large audiences, such as Stranger Things, Supernatural, and American Horror Story (Pennington et al., 2020). Notably, many of these supernatural series include cast members from diverse racial and ethnic backgrounds as the human heroes. Using The Walking Dead as an example, three out of the eight most frequent cast members belong to racial minority groups according to IMDB (“The Walking Dead Full Cast & Crew”, n.d.). The supernatural genre’s high popularity and the nature of its content means that the genre has the potential to promote diversity and inclusion in our society (Ellithorpe, et al., 2018). However, despite the supernatural genre’s potential impact on promoting positive intergroup relationships, there is limited research on the genre and its influence on viewers. Previous studies found that exposure to supernatural genre content with racially diverse heroes and non-human enemies was associated with reduced prejudice toward racial outgroups (Ellithorpe, et al., 2018; Yao et al., 2022a), but more research is needed to replicate and build upon this previous work. One reason that exposure to the supernatural genre may help reduce intergroup conflict lies in the genre’s frequent theme of human heroes fighting non-human villains. It is common in supernatural narratives that human characters are the heroes and non-human characters the outgroup (e.g., The Walking Dead, Sleepy Hollow, Stranger Things), although in some cases the 1 story is presented from the supernatural characters’ points of view which makes the non-human characters the heroes and humans the outgroup (featuring non-human heroes confronting human villains; e.g., True Blood, The Witcher). When the narrative presents human as heroes and non- human (e.g., zombies) as the villains, the viewers are at least temporarily reminded that these non-human creatures are the outgroup, which, in comparison, makes “human” the salient ingroup. Additionally, supernatural genre’s racially diverse cast has been empirically demonstrated to contribute to viewers’ identification with the human category (Ellithorpe et al., 2018). Ellithorpe and colleagues found that viewers identify more with the identity of human when exposed to racially diverse heroes fighting non-human villains compared to all-White heroes fighting non-human villains. This increase in the human identity was associated with reduced prejudice toward Black Americans as well as other U.S. social groups (e.g., Asians). Within the framework of the common ingroup identity model (CIIM; Gaertner & Dovidio, 2000; Gaertner, et al., 1993; Gaertner et al., 2000; Nier et al., 2001), the present research replicates the theoretical mechanism tested in Ellithorpe et al. (2018). In the present research the development and influence of a common ingroup identity (CII; i.e., human identity) and how the CII subsequently affects outgroup prejudice was tracked. Specifically, I tested whether the presence of non-human villains (e.g., zombies), coupled with human heroes being racially homogeneous or diverse, affect outgroup attitudes through strengthening of identity as human. A two-group (exposure to all-White heroes fighting zombies vs. exposure to racially diverse heroes fighting zombies), three-wave longitudinal experiment with a posttest was conducted to investigate the formation of the superordinate human identity through media exposure and how the human identity contributes to reduce racial prejudice over time. According to social identity research, a salient identity in the present study is defined as an identity that is 2 accessible in a situation which is predictive of an individual’s beliefs, attitudes, and behavior (Turner et al., 1987). Self-Categorization and Ingroup Favoritism The theoretical framework of CIIM is an extension of self-categorization theory (SCT; Turner et al., 1987). According to SCT, there are three levels of identity including personal or role, social group, and human or supraordinate (Turner et al., 1987; Turner & Oakes, 1989). A salient personal identity predicts individuals’ behaviors based on the comparison between oneself and other individuals. When group identity is salient, differences between oneself and another individual become less important and group membership is most predictive of evaluative decisions and behaviors through a comparison between in- and outgroup (e.g., race and ethnic groups). According to SCT, and by extension the related social identity theory (SIT, Tajfel & Turner, 1979), ingroup members tend to evaluate their fellow ingroup members in a positive light and judge outgroup members negatively. Such a phenomenon is referred to as ingroup favoritism, which is used by ingroup members to maintain the status quo between the in- and outgroups (Tajfel & Turner, 1979). For example, when racial identity is salient, an individual may attribute positive evaluations to their racial ingroup, and in contrast attribute racist ideologies and negative attitudes toward their racial outgroups. Lastly, the human identity is the superordinate category in which personal and group identities become secondary and, “being a human” becomes dominant in how individuals think and behave. The human identity is more inclusive than group-level identities because all individuals belong to the human category regardless of which social group they identify with. Thus, when the human identity is salient, all other humans become the ingroup and all non-humans are the outgroup. Consistent with the ingroup favoritism concept (Tajfel & Turner, 1979), an individual with a salient human identity 3 should evaluate all other humans (i.e., ingroup) positively, and all non-humans (i.e., outgroup) negatively. Common Ingroup Identity Model Derived from SCT and SIT, the CIIM predicts one possible mechanism through which intergroup prejudice may be altered (Gaertner & Dovidio, 2000; Gaertner, et al., 1993; Gaertner et al., 2000; Nier et al., 2001). This mechanism is recategorization with the common ingroup. The model states that changes in environmental or perceptual cues can change cognitive representations of groups (Gaertner et al., 2000). These cognitive representations can involve seeing people as just individuals without group affiliations (i.e., the personal identity level in SCT), as members of two separate groups (e.g., racial groups like Whites and Blacks), or as members of one larger group which is inclusive of the two subgroup-level separate groups (e.g., the human category is inclusive of all Whites and Blacks; Gaertner et al., 2000). The CIIM predicts that one way to reduce prejudice is through exposure to cues that remind individuals of the larger, more inclusive group (i.e., human category). When the superordinate identity category is salient, individuals may change their cognitive representations of two separate groups into one new and inclusive ingroup (e.g., from Whites as ingroup and racial minorities as outgroup to all humans as ingroup). Thus, intergroup prejudice is expected to decrease as members of the former outgroup are now part of the new, common ingroup. By recategorizing with the common ingroup, individuals are expected to naturally engage in ingroup favoritism to all members of this new, salient ingroup which is inclusive of each subgroup (e.g., Whites vs racial minorities). One study tested cues that may stimulate recategorization with the common ingroup in the context of the supernatural genre. Ellithorpe and colleagues (2018) tested how two different cues impacted strength of identification with the superordinate human category. The narrative 4 villain was manipulated to be human, non-human with human-like appearance (e.g., vampires), or non-human with appearance dissimilar to human (e.g., aliens). They also manipulated the presence of diversity cues for the human character (i.e., a White hero vs. a Black hero). Ellithorpe and colleagues found an interaction between human cues and diversity cues, such that participants who read a story about racially diverse heroes fighting against non-human creatures that are dissimilar to humans reported stronger human identity compared to those who read about White heroes fighting against the same non-human villains. Additionally, evidence of ingroup favoritism within the common ingroup of human was also found, such that as the strength of the human identity increased, prejudice toward Blacks decreased. They also found a generalization effect, where prejudice toward other social groups that were not featured in the story was also decreased. The Present Study The present research was a conceptual replication of Ellithorpe et al. (2018), with a few changes to extend previous research. Ellithorpe and colleagues manipulated two factors in their study – the species of the villain (human vs. non-human1) and race of the hero (Black vs. White). Situated within CIIM, the goal of the original study was to test the situations (e.g., seeing a Black hero) where exposure to non-human villains in the supernatural genre reduces outgroup prejudice through identifying with the superordinate human group. The main theoretical expectation of the original study was that, compared to seeing a White hero fighting non-human villains, participants who saw a minority hero (Black) fighting non-human villains will show stronger identification with the human category, which in turn reduces racial prejudice toward Blacks. 1 Ellithorpe et al. (2018) manipulated the factor into three conditions (human vs. non-human who looks similar to human vs. non-human who looks dissimilar to human) but only human and non-human who looks dissimilar to human produced significant difference in inducing CII. 5 Here, I tested the same prediction as in Ellithorpe et al. (2018). Instead of a one-shot experiment utilized in the original study, I conducted a three-wave, longitudinal controlled experiment with a posttest. The inclusion of a posttest allowed me to compare the two experimental groups after extended media exposure. In addition, three waves of data collection following daily stimuli exposure allowed me to test the mediation proposed by CIIM longitudinally. Second, in my dissertation I simplified the design to hold the species of the villain constant (non-human only) and manipulated the racial identities of the heroes (all-White vs. racially diverse). In this way, I was able to test the mechanism of CIIM in the context of supernatural genre without needing a much larger sample to meet the required statistical power of mediation analysis with longitudinal data. Lastly, the original study manipulated a written transcript of a narrative from the television show Supernatural and used photographs to manipulate both human and diversity cues. Yao et al. (2022a) replicated the study with stimuli being audiovisual scenes taken from the TV series The Walking Dead. Here, I replicated previous research and tested the CIIM process with more controlled audiovisual stimuli. All stimuli were made specifically for the present study, where everything other than the manipulation was held constant. That is, all stimuli videos are identical across conditions except for the human heroes (to be either all White or racially diverse)2. In the present study, the racial identities of human heroes were manipulated by whether the human heroes includes representation from multiple racial and ethnic identities or if all of the human heroes are White. It was expected that the presence of racially diverse heroes will be associated with a stronger human identity. This hypothesized relationship is based on the CIIM’s 2 Gender of the human heroes and the audio narrator was also matched with participant’s gender. For a detailed discussion of stimuli, see “Experimental Stimuli”. 6 predicted variations in the possible cognitive representations of the ingroup (Gaertner & Dovidio, 2000; Gaertner, et al., 1993). As a reminder, the CIIM predicts that individuals can categorize people as individual persons without group affiliation, as members of separate groups, or as members of one common ingroup which is inclusive of subgroups (Gaertner et al., 2000). These different types of category representations are expected to fluctuate the strength of human identity based on cues available. However, the salience of which subgroups are included in the superordinate category of “human” may not always be uniform. Distinctiveness theory predicts that members of the majority identity group are less likely than members of minority identity groups to have group identity salient in neutral contexts (Brewer 2011). Therefore, even if participants are presented with non-human outgroups to remind themselves that they are human, some may fail to extrapolate that the human identity is inclusive of other subgroups (e.g., racial groups) aside from their own. This idea was supported by the findings from Ellithorpe et al. (2018), in that the strength of human identity was higher in the condition with a diverse human fighting non-human enemies compared to a White human fighting non-human enemies. The presence of non-human villains, coupled with reminders of the existence of racial subgroups, should result in cognitive representation of one superordinate human ingroup with racially diverse members. Thus, identity with the common ingroup of humans should be high under these circumstances, and the representation of the ingroup should include racial diversity. H1: Compared to all-White heroes, exposure to racially diverse heroes will be associated with stronger identification with the superordinate human identity group. Previous research on the CIIM has found positive effects of identification with the common ingroup on attitudes toward a subgroup-level outgroup (Ellithorpe et al., 2018; Gaertner et al., 2000; Nier et al., 2001). For example, identification with the human category, a common 7 ingroup identity, has been found to be associated with White participants’ more positive attitudes toward Blacks, which is considered a subgroup-level outgroup to Whites (Ellithorpe et al., 2018). In my dissertation I will test how participants’ attitudes toward their racial outgroups are influenced by the narrative heroes’ racial diversity. According to CIIM, identifying with the common ingroup of humans is the mechanism through which reminders about the human identity affect attitudes toward the subgroup-level outgroup (Gaertner et al., 1993; Gaertner et al., 2000). Thus, I tested the strength of the common ingroup identity of human as a mediator between the manipulation of diversity cues (i.e., human heroes as racially diverse or all White) and prejudiced attitudes. After recategorizing with the common ingroup identity of human, participants are expected to view the minority heroes in the clip (one Black character and one Asian character) as part of their human ingroup and apply ingroup favoritism to their racial groups. As a result, stronger positive attitudes toward Blacks and Asians are expected. H2: Compared to all-White heroes, exposure to racially diverse heroes will be associated with more positive attitudes toward (a) Blacks and (b) Asians, mediated by identification with the superordinate human identity. This newly formed positivity toward Blacks and Asians, who are the target subgroup- level outgroups, is expected to be generalized to other identity groups. This prediction is drawn from previous findings on generalization in the CIIM, as well as research related to secondary transfer effects (Harwood et al., 2011; Joyce & Harwood, 2014). Research in these areas found that the positive attitudes toward the target subgroup-level outgroup are generalized to more positive attitudes toward other subgroup-level outgroups that also belong to the newly formed more inclusive, common ingroup. This generalization effect was also found by Ellithorpe et al. (2018). Therefore, I hypothesize that: 8 H3: Stronger positive attitudes toward (a) Blacks and (b) Asians will be associated with more positive ratings toward other identity groups. According to CIIM, and by extension SIT, a salient CII is required for prejudice reduction to happen (Gaertner & Dovidio, 2000; Gaertner et al., 2000; Turner et al., 1987; Turner & Oakes, 1989). This means changes in attitudes toward minority groups should only occur when the human identity is experimentally manipulated to be salient. This is especially the case for this study because human identity, compared to group-level categories (e.g., race, gender, class), is rarely accessible without cues to make it relevant (Hornsey, 2008). However, what remains unknown is whether repetitive exposures to diverse heroes fighting zombies may produce a lasting effect where participants’ human identity are chronically more accessible than those who watch all-White heroes, and whether these participants will consequently report weaker racial prejudice toward minority groups even when their human identity is not experimentally made salient. I tested this research question in the posttest which was disseminated one day after the last stimuli exposure. RQ1: Compared to all-White heroes, will participants who are exposed to racially diverse heroes report stronger human identity at the posttest? To test whether media exposure might influence the chronic accessibility of participants’ attitudes toward a target minority group, a jury simulation task was included in the posttest (Bond, 2021). Because of recent increases in discrimination against Asians in the U.S. (e.g., Bekiempis, 2021; Chiu, 2020), Asians served as the example target group for this test. Participants were to rate how severe the punishment should be for a hate crime targeting Asian people. Outgroup prejudice should predict sentencing severity, such that stronger prejudice will be associated with weaker intended sentencing severity. Participants who are exposed to diverse 9 heroes fighting zombies are expected to show stronger human identity and more positive attitudes toward minority groups, which indicates weaker prejudice, which means that they will select more severe punishment for the Asian hate crime compared to participants in the all-White heroes condition. However, as mentioned earlier it remains unknown whether the CII mechanism is effective in prejudice reduction longitudinally after the immediate manipulation induction of human identity. RQ2: Compared to the condition with all-White heroes, will participants who are exposed to racially diverse heroes report more severe punishment for those who perpetrated an Asian hate crime? RQ3: Will the association mentioned in RQ2 be mediated by identification with the human identity? 10 CHAPTER 2 METHOD Participants & Procedure The present longitudinal experiment involved a four-wave procedure with one day between each wave. The one-day interval is a reasonable choice because previous research on longitudinal media effects has found significant results when setting the between-wave intervals to be one day (Knobloch-Westerwick & Crane, 2012). All surveys were administered online. Participants were allowed to complete the survey on their own time within the bounds of each wave and on their preferred devices. This design was to increase ecological validity by allowing participants to potentially use their typical media-use device at a time when they typically consume media. Additionally, this design was intended to reduce attrition rate. In Wave 1’s (W1) survey, participants were directed to watch a stimulus video and then answer a series of questions (see Table 1 for detailed measures in each wave). Wave 2 (W2) and 3’s (W3) survey questions were identical. In each survey participants watched a video followed by questions on the key measures. In Wave 4’s (W4) survey (i.e., the posttest), participants responded to all key measures as well as completed a jury simulation task without watching a video. There was a manipulation check question in each of the W1-3 surveys and an attention check question in all four surveys. See Table 1 for an overview of the longitudinal experiment design and measures (measures will be discussed in detail under “Measures”). The present study was approved by the Institutional Review Board at Michigan State University. 11 Table 1 Overview of the Longitudinal Experiment Design Wave Session Measures Strength of human identity Attitudes toward identity groups (including Stimuli exposure, Blacks and Asians) measures of key W1 Demographics variables, and Auxiliary variable demographics Manipulation check Attention check Strength of human identity Stimuli exposure, Attitudes toward identity groups (including W2 measures of key Blacks and Asians) variables Manipulation check Attention check Stimuli exposure, W3 measures of key Same as Day 3 variables Jury simulation task Attitudes toward minority groups (including W4 Posttest Blacks and Asians) Strength of human identity An a priori power analysis was conducted to determine a sample size sufficient for the mediation analysis, where bias-corrected confidence intervals of the mediation coefficient were computed with 1000 bootstrap samples and 100 Monte Carlo simulation replications. The statistical power was determined by the percentage of the 100 Monte Carlo simulation replications where the mediation was significant from the bootstrap samples. Based on the power analysis, to replicate the effect sizes (e.g., standardized coefficients) of Ellithorpe et al. (2018), a minimum sample size of 400 is required at W3 to find a statistically significant mediation with a statistical power of roughly .8. A nonrepresentative U.S. adult sample was obtained through a private survey research company Dynata (Dynata LLC: Better Business Bureau® Profile, n.d.), who planned the initial 12 sample size based on the typical attrition rate of the platform. Gender and age were balanced as quota in the initial wave but were subjected to natural fallout in subsequent waves. Participants were randomly assigned to one of the two conditions and stayed in the same condition across all four waves. Of the initial 3,081 participants who started the survey in W1, 892 were removed from the survey due to answering the attention check question incorrectly and their responses were not recorded. All of the remaining 2,189 participants were invited via messages from the Dynata system to participate W2. Of these invited participants, 339 started the W2 survey and 129 failed the attention check question, which left a total of 210 participants who completed the survey in W2. Participants who completed W1 and W2 as well as those who completed W1 but not W2 were invited for W3, which acted effectively to reduce attrition rate. Such a decision was also justified as the experimental induction of W3’s survey was not dependent on the completion of W2’s survey. A total of 303 participants started the survey in W3 and 270 completed the survey with the correct answer to the attention check question. For W4, participants who participated at least two out of the first three waves were invited (n=335) and 257 completed the W4 survey. Overall, a total of 143 participants completed all of the first three waves of data collection which is far less than the desired sample size of 400 from the power analysis. To obtain a stronger statistical power for data analysis, the final sample for analysis includes the four-wave data from the 335 participants who completed at least two out of the first three waves (age ranged from 18 to 87; Mage= 60.58; SDage=17.05; 35% were self-identified women; racial breakdown: 85% White, 5% Black or African American, 4% Asian, 2% Latino/a or Hispanic origin, and 4% interracial). 13 Design The present longitudinal controlled experiment used a one factor (diversity cues: present vs. absent) between-subjects design, using random assignment to determine participants’ condition. Diversity cues was manipulated by the race of the heroes in the video, which is present when the heroes represent diverse racial groups and absent when all heroes in the video are White. To ensure the ease for participants to recognize the presence and absence of the diversity cues, photographs of all heroes in the video were provided on the screen immediately before participants watched the video. Experimental Stimuli Each stimulus video of the present study included a fighting scene between human heroes and zombies. The audiovisual materials of the stimulus videos were made to be constant across conditions except for the manipulation (i.e., racial identities of the human heroes). The script of the stimuli audio was written by a professional writer, where fights between human heroes and zombies were highlighted in each wave’s story. Two professional voiceover performers (a woman and a man, to match participant gender) were then hired to read the scripts to create the audio part of the stimuli. The visual part of the stimuli included different combinations of pictures representing each segment of the script. The racial identities of the human heroes were manipulated by their photographs to either include three White characters or three racially diverse characters (i.e., one White character, one Black character, and one Asian character). The final video stimuli were made by matching the visual part to the audio part, so that the visual material meaningfully represents each segment of the audio material. The content of the video stimuli depicts three friends that defended their homebase from a zombie attack, who then encountered more zombie attacks on their way to the local abandoned hospital for medical 14 supplies. The stimuli videos of W1 to W3 shared a coherent storyline to mimic people’s media exposure in their everyday lives (i.e., environment realism). The main characters’ names were the same as in Ellithorpe et al.’s original study (2018) and gender matched for each participant. The female version of main characters were Ashley, Sarah, and Jessica, and the male version of main characters were Michael, Christopher, and Andrew. Overall, video stimuli were identical across conditions except for (1) the race of the human heroes which is the experimental manipulation, (2) gender and names of the human heroes which were used to match participant’s gender, and (3) gender of the narrators which was also used to match participant’s gender. All video stimuli were about five minutes long (range= 4:05 to 4:47). Human hero photographs were pilot tested. The manipulation of the diversity cues was operationalized by the photographs of the human heroes. There are one White character, one Black character, and one Asian character in the diverse hero condition and three White characters in the White hero condition. Because hero photographs were designed to match participant’s gender, there were also a female version and a male version. With the White character in the diverse hero condition also used in the White hero condition, a total of 10 photographs were needed in the final stimuli (three female diverse heroes, three male diverse heroes, two additional White female heroes, and two additional White male heroes). Three possible photographs were pilot tested for each character. The pool of pilot tested photographs were downloaded from a stock image website (istockphoto.com). Ellithorpe et al.’s (2018) original research has one main character who “regardless of sex or race, was a young adult wearing a black leather jacket, standing against a plain background and looking at the camera” (p. 510). To be consistent with the original research, I searched “young adult in black leather jacket” on the website and then subset the results by gender and race with the “refine” function. 15 Consistent with the original research, all photographs selected for the pilot test included one young adult wearing a black leather jacket, standing against a plain background, and looking at the camera. Participants for the pilot test (n=195; age ranged from 18 to 38; Mage=20.19; SDage=1.95; 74% were self-identified women; racial breakdown: 66% White, 5% Black or African American, 17% Asian, 9% interracial, and 3% other) were undergraduate students from a large mid- Western university, who participated in the online survey for course credit. In the survey, female participants were directed to evaluate all female photographs and male participants were directed to evaluate all male photographs. The order of the photographs that were rated was randomized. After viewing each photograph, participants were asked to select the racial group that the person is “most likely to be a member of” (options include White, Black or African American, American Indian or Alaska Native, Asian, Native Hawaiian or Pacific Islander, Hispanic or Latino/a, Middle Eastern, and other; the order of the options was randomized). Participants were allowed to select one or more racial groups for the person in each photograph. Participants also rated likeability and attractiveness of the person in the photograph. Additionally, participants were asked whether they have seen the person in the photograph before to potentially exclude a well-known individual from the final stimuli. The primary purpose of the photographs was to present the person in it with clear racial identity. For example, a White character needs to be easily recognized as White and only White. The final set of photographs were selected based on the percentage of correct recognition of the person’s race based on the condition he or she is in (range=75% to 99%). Although it was not possible to select a set of photographs with clear racial identities as well as same likeability and attractiveness, effort was made to balance these perceived traits across characters. 16 Measures See Table 2 for descriptive statistics. Table 2 Descriptive Statistics Variable Wave Range M SD Cronbach α Strength of W1 [1, 7] 5.70 2.08 - human identity W2 6.00 1.86 - W3 5.90 1.96 - W4 6.09 1.84 - Attitudes W1 [0, 100] 70.93 26.69 - toward Blacks W2 72.13 25.26 - W3 71.25 27.08 - W4 71.35 27.50 - Attitudes W1 75.16 23.27 - toward Asians W2 77.61 21.56 - W3 76.47 23.29 - W4 77.06 23.53 - Attitudes W1 [0, 100] 65.08 24.95 .90 toward other W2 65.91 25.07 .92 identity groups W3 65.91 26.25 .92 (no Blacks) W4 66.38 26.58 .93 Attitudes W1 63.80 25.14 .90 toward other W2 64.78 25.43 .92 identity groups W3 64.33 26.67 .92 (no Asians) W4 64.86 27.11 .93 Jury W4 [1, 5] 3.21 1.26 - simulation task Auxiliary - [-5, 5] 2.35 1.92 .71 variable 17 Strength of the superordinate human identity was measured by how closely participants feel to the identity group of humans, which was operationalized through a modified version of the Inclusion of theIingroup in the Self Scale (Gómez, Tropp, & Fernández, 2011; Tropp & Wright, 2001), which was also used as the measure of human identity strength in Ellithorpe et al. (2018). Participants reported how close or distant they identify with the human category with seven pairs of circles that begin completely apart and gradually come closer until they nearly completely overlap (range=1 to 7). A higher number means more overlap between the pair of circles, which indicates stronger identification with the human category. Attitude toward Identity Groups was measured. Participants were asked to rate how they feel about other racial and ethnic groups on “feeling thermometers” ranging from zero (strongly dislike) to 100 (strongly like). These groups include Black, American Indian and Alaska Native, Asian, Native Hawaiian and other pacific islander, and Latino or Hispanic Origin. Participants’ scores of attitudes toward Blacks were their rating of the Black racial group, and participants’ scores of attitudes toward Asians were their rating of the Asian racial group. Black participants’ ratings about their own racial group were excluded from analyses, and Asian participants’ ratings about the Asian group were excluded from analyses. The average of participants’ ratings of all groups were used as their attitudes toward other identity groups score. To avoid using the same variable twice in one model, for analyses with Blacks as the target group attitudes toward other identity groups score was the average of a participant’s ratings of all measured identity groups except for Blacks. Similarly, for analyses with Asians as the target group attitudes toward other identity groups score was the average of all measured identity groups except for Asians. Additionally, each participant’s rating of their own racial group was excluded from analyses. For example, an American Indian participant’s 18 attitudes toward minority groups score in the Asian target group model was the average of their responses to the measured groups except for American Indians (or Alaska Native) and Asians. In the posttest, participants completed a jury simulation task and were presented with three narratives of fictional court cases (Bond, 2021). Two cases were distraction items, but the third case describes a perpetrator of an Asian hate crime who is found guilty. The order of the three items was randomized. Participants were asked to determine the most appropriate punishment to accompany the guilty verdict. Participants were presented with five sentencing options that ranged in severity from 1 (a warning) to 5 (a significant fine accompanied by probation). A higher score will represent more severe punishments for the defendant found guilty of anti-Asian discrimination, which indicates more positive attitudes toward the Asian group. Due to the potential sensitivity of the question, all Asian participants skipped this question by survey design. After viewing the stimulus video in the survey (W1-3), participants completed the manipulation check questions and were asked to recall the hero races in the video to be one of the following: a group of three White male adults, three White female adults, three male adults with different racial identities, or three female adults with different racial identities. In the final sample, 64% of participants answered the manipulation check question correctly in W1, 88% in W2, and 88% in W3. Full information maximum likelihood (FIML; Enders & Bandalos, 2001) was used to address missing data in the present research which works better when an auxiliary variable3 is controlled in the model (Graham, 2003). Because participants of the current study are adults 3 An auxiliary variable is a variable that is “not included in the analysis (as part of the key variables), but correlated with a variable with missing values and/or related to its missingness” (Jacobsen et al., 2017). 19 from the general population, the missingness of the data was expected to correlate with individual’s conscientiousness. That is, participants in my study who scored high on the measures of conscientiousness were expected to be less likely to skip one or more days of the survey assignments. Thus, the four-item conscientiousness subscale from the Mini-IPIP Inventory (Donnellan et al., 2006) was measured and used as an auxiliary variable in the longitudinal SEM models. A participant’s conscientiousness score (i.e., the auxiliary variable) was the average of the four items. An example item is “I get chores done right away.” 20 CHAPTER 3 STATISTICAL ANALYSIS All hypotheses and research questions were tested using structural equation modeling (SEM) with the Lavaan package (Version 06-3; Rosseel, 2012) in R (Version 4.2.1). Two SEM models were estimated for the present research. Specifically, a cross-lagged panel model (CLPM) was estimated with data from all four waves for H1-3, and a simple mediation model was estimated for the research questions. Both of these models have two tested versions, one with Blacks as the target group and one with Asians as the target group. For both models, because the local direct and indirect effects are of interest rather than global model fit, saturated models were used and as a result there was no global model fit indices to be reported. Additionally, because within-person changes for each participant over time (i.e., coefficients for each participant: the random effects) is not of interest in the present study, both models were estimated with fixed effects only. Missing data was addressed with FIML within model building, which is a widely used method and generally preferred over multiple imputation whenever possible (Larsen, 2011). Additionally, participants’ self-reported race was included in all analyses as a covariate. To have all variables included in the models on similar measurement scales, scores on attitudes toward Blacks, attitudes toward Asians, and attitudes toward other identity groups were recoded into between zero and ten (i.e., divided the original scores [range: 0 to 100] by ten). 21 Figure 1 Path Model H1-3 were tested with a CLPM with data from W1 to W4 (see Figure for the path model). As seen in the Figure, the top row represents exposure condition in W1 through W3. Because the experimental conditions are constant, the path coefficient between exposure condition at W1 and W2 and the path coefficient between exposure condition at W2 and W3 were restricted as 1.00. The second row from the top represents the four waves’ measure of the mediator –strength of the human identity (H2). The third row from the top represents the four measures of attitudes toward Blacks, which is the dependent variable predicted in H2a. To test H2b a different model was estimated with attitudes toward Blacks replaced by attitudes toward Asians. The bottom row represents four measures of attitudes toward other groups. As a reminder, when attitudes toward Blacks were included in the model, attitudes toward other identity groups were the mean of all measured minority groups except for Blacks. Similarly, when attitudes toward Asians were in the 22 model, attitudes toward other identity groups were the mean of all measured minority groups except for Asians. All parallel variables (e.g., all variables from W1 in the Figure) were allowed to covary. All between-wave paths pointing to the exposure condition variables were fixed to be zero, as experimental conditions are not expected to be predicted by any of these variables. All other between-wave paths were included in the model due to the specification of a saturated model. Lastly, because FIML was used to address missing data, an auxiliary variable was included in the regression estimations of every endogenous variable within the SEM model (Enders & Bandalos, 2001). As mentioned earlier, an auxiliary variable is one that is not part of the key measures but included in the statistical model because it is expected to correlate with the missingness of the data. The goal of including an auxiliary variable is for FIML algorithm to work better. In both the Blacks as target group model and the Asians as target group model, H1 was tested by whether path a1 was statistically significant. As recommended by Hayes (2018), the mediation predicted in H2 was tested by multiplying the path coefficients a1 and b2. H3 was tested by whether c3 was statistically significant. Using the bootstrapping technique to test statistical significance of path coefficients in longitudinal SEM is less straightforward than in the typical case, and attempt to do so resulted in convergence issues (after 80 iterations) in the present CLPM models. Fortunately, because the CLPM used only observed variables to specify the CLPM models, all path results are completely reliable regardless of whether the model converges 4. Thus, instead of bootstrap samples, I used Monte Carlo simulations (20,000 4 An SEM model is said to have converged when the residual matrix (i.e., a matrix of differences between the observed parameters and estimated parameters) cannot be minimized any further based on the iterative model estimation process (Kline, 2016). When using only observed variables in an SEM model, all path coefficients are computed with observations, thus there is no residual matrix available. In other words, if an SEM model only includes observed variables, all path coefficients remain the same even when the model does not converge. 23 repetitions) to produce confidence intervals (CI) which is a recommended alternative to bootstrap (Imai et al., 2010; Preacher & Selig, 2012; see also Long et al., 2021, for an example). RQ1-3 were tested by a simple mediation model using W4 data only, which included the exposure condition as the independent variable, the strength of human identity as the mediator, and responses to the jury simulation task (i.e., attitudes toward Asians) as the dependent variable. This model is again a saturated model estimated with fixed effects only. Reported regression coefficients are unstandardized, and 95% CIs are reported with 5,000 bootstrap samples and bias-corrected estimators. Missing data was addressed with FIML and participants’ race was included as a covariate. RQ1 was tested by whether the path between exposure condition and the strength of human identity is statistically significant. RQ2 was tested by whether the path between the exposure condition and attitudes toward Asians is statistically significant. The mediation was tested by multiplying the previously mentioned two paths (see Hayes, 2018), which was tested to answer RQ3. To cross-validate the results to RQ1 and RQ2 with a different method, two one-way between-subject ANCOVAs were used. One ANCOVA tested whether strength of human identity would be different between exposure conditions at posttest (W4), controlling for participant’s race. Another ANCOVA tested whether responses to the Asian hate crime would be different between exposure conditions at posttest (W4), controlling for participant’s race. 24 CHAPTER 4 RESULTS See Table 3 for a summary of results for H1-3. H1 predicted that exposure to racially diverse heroes will be associated with stronger identification with the superordinate human identity compared to exposure to all-White heroes. As seen in Table 3, this hypothesis was not supported by the data when the target group is Blacks or Asians. H2 predicted that exposure to racially diverse heroes will be associated with more positive attitudes toward (a) Blacks and (b) Asians, and this relationship will be mediated by increased human identity. Regardless of Blacks or Asians as the target group, strength of human identity did not mediate the relationship between diversity cues and attitudes toward the target outgroup. Additionally, an increase in strength of human identity was not significantly associated with an increase in the positivity of attitudes toward the outgroup. H2 was rejected. H3 predicted an association between attitudes toward the target outgroup (Blacks or Asians) and attitudes toward other identity groups. This hypothesis was also not supported by data for Blacks and Asians target groups. For both target groups, more positive attitudes toward the target outgroup did not significantly predict more positive general attitudes to other identity groups. Overall, none of the H1-3 was supported by data. 25 Table 3 Hypothesized Direct and Indirect Effects Coeff. H1 Diversity cues(W1)→strength of human .09[-.51, .86] identity(W2) Blacks H2 Strength of human identity(W2)→Attitudes toward -.004[-.59, .59] as Blacks(W3) target Diversity cues(W1)→strength of human -.000[-.10, .10] group identity(W2)→Attitudes toward Blacks(W3) H3 Attitudes toward Blacks(W3) →Attitudes toward .06[-.58, .81] other identity groups(W4) H1 Diversity cues(W1)→strength of human .09[-.52, .88] identity(W2) Asians H2 Strength of human identity(W2)→Attitudes toward .072[-.41,.68] as Asians(W3) target Diversity cues(W1)→strength of human .006[-.08, .11] group identity(W2)→Attitudes toward Asians(W3) H3 Attitudes toward Asians(W3) →Attitudes toward .03[-.70, .80] other identity groups(W4) Note. Coeff. = Coefficients. In diversity cues, diverse hero = 1. Coefficients are unstandardized. 95% confidence intervals were calculated with 20,000 Monte Carlo repetitions, which is presented in the square bracket next to each coefficient. A simple mediation model was tested to answer RQ1-3. RQ1 asked whether watching diverse heroes fighting zombies would lead to stronger human identity at the posttest compared to watching White heroes fighting zombies. Results indicated that the increase in human identity at the posttest after watching diverse heroes instead of White heroes for three days was not statistically significant (b=.33, 95% CI=[-.12, .81]). A one-way, between-subject ANCOVA produced the same finding, that the two conditions were not significantly different in identification with the human category at the posttest (F(1, 252)=2.00, p=.16), controlling for participants’ race. RQ2 asked whether watching diverse heroes fighting zombies for three days would lead to more positive attitudes toward racial outgroups at the posttest, which was operationalized by a question asking about the severity of intended punishment for an Asian hate 26 crime. Results showed that watching diverse heroes did not lead to more severe intended punishment in question (b=-.20, 95% CI=[-.54, .13]). Such a finding was consistent with the ANCOVA test, that the intended punishment for the Asian hate crime at the posttest did not significantly differ by experimental conditions (F(1, 218)=1.35, p=.29). RQ3 asked whether the relationship between diversity cues and responses to Asian hate crimes would be mediated by human identity at the posttest. The simple mediation between diversity cues, strength of human identity, and response to the Asian hate crime was not statistically significant (b=-.002, 95% CI=[-.01, .11]). Additionally, the path between the strength of human identity at the posttest and the response to Asian hate crime was statistically nonsignificant (b=.07, 95% CI=[-.03, .17]). 27 CHAPTER 5 DISCUSSION Intergroup prejudice has been an important issue in our society. Empirical research has found that identifying with the superordinate human category through watching certain types of media content may be an effective mechanism to reduce intergroup prejudice (e.g., Ellithorpe et al., 2018). For example, supernatural genre with human heroes and non-human villains may temporarily help viewers strengthen their human identity and recognize that people with different racial and ethnic identities are all part of their human ingroup. Both Ellithorpe et al.’s study and a replication study, Yao et al. (2022a), have found that the strength of human identity mediates the relationship between watching racially diverse human heroes fighting against non-human villains and more positive attitudes toward racial minority groups. Despite the consistent findings, both of these studies tested mediation with a single-shot experiment. Mediation analysis with cross- sectional data has its merit but this analysis could also be done with more statistical rigor such as using longitudinal data (Maxwell & Cole, 2007). My dissertation replicated Ellithorpe et al. with a longitudinal experiment and the mediation proposed by CIIM was tested with a statistical model that explicitly considers time (i.e., CLPM). Overall, no significant mediation was found in the longitudinal setting, though I believe this set of results offers unprecedented understanding about the time scale of the salience of human identity and how it may be used to promote positive interracial relationships. The mediation proposed in CIIM (H2) was tested with three waves of data across three days. As seen in Table 3, diversity cues at W1 did not significantly affect strength of human identity at W2, and these two variables were measured with roughly one day apart. Similarly, strength of human identity at W2 did not significantly affect attitudes toward the target outgroup 28 at W3 (i.e., Blacks or Asians). These variables were also measured with roughly one day apart. Additionally, the simple mediation between the three variables was found to be nonsignificant for both target groups. This set of findings is consistent with neither Ellithorpe et al.’s (2018) original research nor Yao et al.’s (2022a) replication research. There are a few ways to interpret these results. These nonsignificant findings could mean that the previous support for the CII mechanism was unstable and failed to hold when changes in time is explicitly considered. Or, the nonsignificant findings could be due to lack of statistical power (i.e., an a priori power analysis indicated a desired sample size of 400 but the final sample size was 335). If that is the reason of the non-significance of the present findings, then future research needs to replicate this study with a sufficient sample size. Lastly, it is also possible that the CII (e.g., human identity) is a valid mechanism to reduce prejudice, only the salience of the human identity - a specific kind of CII – does not last very long. Indeed, previous literature has argued that the human identity is not a natural category, and people may temporarily show salient human identity when prompted but eventually will return to their lower-level identity categories (Hornsey, 2008). If CII is a valid mechanism but with a short-term salience period of the human identity, then the present data should offer support when testing the hypotheses within each wave. In each of the surveys from W1 to W3, the strength of human identity is measured immediately after stimuli exposure which is then immediately followed by measures of outgroup attitudes. Post hoc analyses confirmed my prediction, that when measuring the effects of human identity right after it is made salient (as Ellithorpe et al. did in their research), significant mediation effects were found. As seen in Table 4, five out of the six sets of analyses demonstrated evidence of CIIM and were consistent with Ellithorpe et al.’s findings. Overall, the findings of the present study showed that CII is an effective mechanism on prejudice reduction but the salience of the 29 human identity may be short-lived. Empirical evidence in this study demonstrated that salience of the human identity does not last as long as one day. But how long exactly can the human identity remain salient? Future research should answer this important question with novel research methods in addition to varied frequency and dosage of media exposure. 30 Table 4 Mediation Effects within Each Wave Coeff. Diversity cues→Strength of human Blacks .10[.01, .27] identity→Attitudes toward Blacks as Diversity cues→Strength of human target identity→Attitudes toward Blacks→Attitudes .08[.004, .22] group toward other identity groups W1 Diversity cues→Strength of human Asians .13[.01, .30] identity→Attitudes toward Asians as Diversity cues→Strength of human target .12[.01, .30] group identity→Attitudes toward Asians→Attitudes toward other identity groups Diversity cues→Strength of human Blacks .08[-.01, .32] identity→Attitudes toward Blacks as Diversity cues→Strength of human target .07[-.01, .29] group identity→Attitudes toward Blacks→Attitudes toward other identity groups W2 Diversity cues→Strength of human Asians .12[.01, .35] identity→Attitudes toward Asians as target Diversity cues→Strength of human .13[.01, .38] group identity→Attitudes toward Asians→Attitudes toward other identity groups Diversity cues→Strength of human Blacks .14[.02, .37] identity→Attitudes toward Blacks as Diversity cues→Strength of human target .12[.02, .33] group identity→Attitudes toward Blacks→Attitudes toward other identity groups W3 Diversity cues→Strength of human Asians .20[.02, .46] identity→Attitudes toward Asians as Diversity cues→Strength of human target .21[.03, .49] group identity→Attitudes toward Asians→Attitudes toward other identity groups Note. Coeff. = coefficients. Saturated models were estimated, therefore there was no model fit indices to be reported. Coefficients presented here are unstandardized, bias-corrected, and with 5,000 bootstrap samples. 95% CI is presented in the square bracket next to each coefficient. Coefficients with confidence intervals that do not contain zero are bolded. Notably, the present results align with findings from priming effects research. Overall the current data implies that the salience of human identity can last at least several seconds or several 31 minutes (e.g., the time lag between stimuli exposure in W1 and responding to the human identity question in W1) but probably not as long as up to one day (e.g., the time lag between stimuli exposure in W1 and responding to the human identity question in W2). This is consistent with previous priming research which has found that priming effects lasted no longer than one hour (e.g., Roskos-Ewoldsen et al., 2007). Additionally, in the present study the hypothesized paths were non-significant when the lag was one day but were likely to be significant when the lag was several seconds to several minutes. These findings are consistent with the notion that the effects of media primes do fade with time (Grant & Logan, 1993; Ewoldsen & Rhodes, 2019). However, as mentioned earlier the current data does not offer enough information on how long exactly the human identity remains accessible after it is experimentally primed. Future research is recommended to test the focal phenomenon from a theoretical perspective of media priming effects. In the present research, a more direct test of whether the human identity can be chronically accessible lies in the research questions. At the posttest, I tested whether participants’ human identity would remain salient after three days of diversity exposure but with no experimental induction of this identity during the fourth day’s survey. The results showed that neither the human identity nor outgroup attitudes were significantly different at the posttest, and the effect of exposure condition on outgroup attitudes was not mediated by the human identity. These results provided further evidence that the salience of human identity may not be long- lasting. One caveat in the findings for the research questions is the non-significant relationship between strength of human identity and outgroup attitudes (i.e., intended severity for the punishment of an Asian hate crime). The positive relationship between CII and outgroup 32 attitudes has been validated extensively in previous research, so this non-significant finding could indicate some sample-specific issues such as response quality of certain measures. To test the robustness of this finding, I did more post hoc analyses, replacing the Asian hate crime measure with the warmth thermometer scores for Asians and Blacks (see Table 5 for results). In both the model with Asians as target group and the one with Blacks as target group, the strength of human identity was positively and significantly related to outgroup attitudes, consistent with prior findings. Though strength of human identity as well as outgroup attitudes at the posttest showed no difference between experimental conditions. Table 5 Post Hoc Analyses for RQ 1-3 with Data from the Posttest (W4) Coeff. Diversity cues→strength of human identity .32[-.12, .81] Blacks Strength of human identity→Attitudes toward Blacks .27[.04, .49] as Diversity cues→Attitudes toward Blacks .05[-.06, .77] target .09[-.02, .30] group Diversity cues→strength of human identity→Attitudes toward Blacks Diversity cues→strength of human identity .33[-.12, .81] Asians Strength of human identity→Attitudes toward Asians .33[.13, .53] as Diversity cues →Attitudes toward Asians -.71[-1.29,.12] target .11[-.03, .31] group Diversity cues→strength of human identity→Attitudes toward Asians Note. n=256. Coeff. = coefficients. Saturated models were estimated, therefore there was no model fit indices to be reported. Coefficients presented here are unstandardized, bias-corrected, and with 5,000 bootstrap samples. 95% CI is presented in the square bracket next to each coefficient. Coefficients with confidence intervals that do not contain zero are bolded. Limitations and Future Research One potential limitation of this research lies in the duration between each pair of waves. There is a lack of research in the focal context to provide useful guidance about the proper length of between-wave interval. I chose one day to be the interval because one study in media 33 psychology has found significant effects when setting the time between two waves to be this long (Knobloch-Westerwick & Crane, 2012). However, it remains unknown whether a one-day interval is proper when testing the CII processes. The current results showed that between-wave intervals could be a critical aspect in this line of research. Indeed, when testing how human identity influences outgroup attitudes one day later, no significant relationships were found. But strong and significant relationships were found when the responses of outgroup attitudes were collected right after the human identity was made salient. Future longitudinal research in this area is strongly recommended to continue testing how long a salient human identity would last and, as a result, an appropriate duration for the between-wave intervals. Another limitation of the present research lies in the lack of statistical power. The a priori power analysis indicated the desired sample size (i.e., participants who completed all first three waves) to replicate Ellithorpe et al.’s findings to be 400. In the final sample, a total of 143 participants completed the first three waves which is much lower than the desired 400 sample size5. To obtain stronger statistical power than a sample size of 143, I used participants who completed at least two out of the first three waves (n=335). Data quality may be threatened by including participants who skipped a wave of data collection, though the missing data issue was addressed by FIML. However, even with the loosened threshold, the final sample size (n=335) was still smaller than the desired 400. It is strongly recommended that future research replicates the present study with enough statical power. Finally, even though the present research did not show human identity to be chronically salient after repetitive media exposure, it does not mean that the human identity cannot be chronically salient. Previous research has demonstrated that chronic accessibility of concepts 5 I was constrained by funding and as a result was not able to get more participants for this study. 34 increases with the higher frequency of exposure (Roskos-Ewoldsen & Fazio, 1992; Wyer, 2004). Extensive research has also found that long-term exposure to a certain topic may increase accessibility of attitudes toward the topic (Arendt, 2010; Arendt & Brantner, 2015). Thus it is possible that the short salience of the human identity in the present research is due to low frequency and dosage of the diversity exposure in the video stimuli. After all, a total of 15- minute stimuli exposure across three days seems to be much shorter compared to the media consumption of a regular viewer of the supernatural genre. That said, where the tipping point (i.e., how frequent and how long the media exposure is enough) lies for the human identity to be chronically salient is yet to be discovered. Conclusion As intergroup conflict continues to be a major issue in today’s society, communication researchers have validated various mechanisms with the potential to promote positive intergroup relationships (e.g., Ellithorpe et al., 2022; Joyce & Harwood, 2014; Schiappa et al., 2005). The CII, especially the human identity, has been demonstrated to be an effective tool to combat negative outgroup attitudes. The present study replicated and extended one of the first studies (Ellithorpe et al., 2018) examining the effects of exposure to racial diversity in media on viewers’ social group attitudes. The longitudinal experimental design of the present research allowed the mediation proposed by CIIM to be tested with more statistical rigor, which also allowed the chronic accessibility of the human category to be tested. Although the results were non-significant for the longitudinal mediation, the present findings offered important implications and raised interesting questions for future research to study. Mass media content with human heroes and non-human villains (e.g., supernatural genre; sci-fi genre) has drawn vast popularity in recent years, and I hope my dissertation will provide information for scientific 35 researchers and media content producers to effectively promote positive interracial relationships through the CII processes. 36 REFERENCES 37 REFERENCES Arendt, F. (2010). Cultivation effects of a newspaper on reality estimates and explicit and implicit attitudes. Journal of Media Psychology: Theories, Methods, and Applications, 22(4), 147–159. https://doi.org/10.1027/1864-1105/a000020 Arendt, F., & Brantner, C. (2015). Toward an Implicit Cognition Account of Attribute Agenda Setting. International Journal Of Communication, 9, 22. Retrieved from https://ijoc.org/index.php/ijoc/article/view/4001 Bekiempis, V. (2021, February 20). ‘Alarming surge’ in anti-Asian violence across US terrifies community. The Guardian. Retrieved from https://www.theguardian.com/us- news/2021/feb/20/anti-asian-violence-us-bigotry, Accessed 24 July 2021. Bond, B. J. (2021). The Development and Influence of Parasocial Relationships With Television Characters: A Longitudinal Experimental Test of Prejudice Reduction Through Parasocial Contact. Communication Research, 48(4), 573– 593. https://doi.org/10.1177/0093650219900632 Brewer, M. B. (2011). Optimal distinctiveness theory: Its history and development. Handbook of theories of social psychology, 2, 81-98. Chiu, A. (2020, March 20). Trump has no qualms about calling the coronavirus the ‘Chinese Virus.’ That’s a dangerous attitude, experts say. The Washington Post. Retrieved from https://www.washingtonpost.com/nation/2020/03/20/coronavirus-trump-chinese- virus/. Accessed 24 July 2021. Dynata LLC: Better Business Bureau® Profile. (n.d.). Retrieved October 05, 2020, from https://www.bbb.org/us/ny/new-york/profile/internet-service/dynata-llc-0121-91228 Doosje, B., Ellemers, N., & Spears, R. (1995). Perceived intragroup variability as a function of group status and identification. Journal of experimental social psychology, 31(5), 410- 436. https://doi.org/10.1006/jesp.1995.1018 Donnellan, M. B., Oswald, F. L., Baird, B. M., & Lucas, R. E. (2006). The Mini-IPIP Scales: Tiny-yet-effective measures of the Big Five Factors of Personality. Psychological Assessment, 18(2), 192–203. https://doi.org/10.1037/1040-3590.18.2.192 Ellithorpe, M. E., Holt, L. F., & Ewoldsen, D. R. (2022). Would they save me, too? Victim race recall when the hero is Black vs. White and its influence on expectations of reciprocity. Media Psychology, 1-17. https://doi.org/10.1080/15213269.2022.2078842 Ellithorpe, M. E., Ewoldsen, D. R., & Porreca, K. (2018). Die, foul creature! How the supernatural genre affects attitudes toward outgroups through strength of human identity. Communication Research, 45(4), 502-524. https://doi.org/10.1177/0093650215609674 38 Enders, C. K., & Bandalos, D. L. (2001, 2001/07/01). The Relative Performance of Full Information Maximum Likelihood Estimation for Missing Data in Structural Equation Models. Structural Equation Modeling: A Multidisciplinary Journal, 8(3), 430-457. https://doi.org/10.1207/S15328007SEM0803_5 Ewoldsen, D. R., & Rhodes, N. (2019). Media priming and accessibility. In Oliver M. B., Raney A. A., & Bryant J. (Eds.), Media effects (4th ed., p. 83-99). Routledge. Gaertner, S. L., Dovidio, J. F., Nier, J. A., Banker, B. S., Ward, C. M., Houlette, M., & Loux, S. (2000). The common ingroup identity model for reducing intergroup bias: Progress and challenges. In D. Capozza & R. Brown (Eds.), Social identity processes: Trends in theory and research (pp. 133-148). Thousand Oaks, CA: Sage. Gaertner, S. L., & Dovidio, J. F. (2000). Reducing intergroup bias: The common ingroup identity model. Psychology Press. Gerbner, G. (1977). Television: The new state religion? ETC: A Review of General Semantics, 34, 145–150. Gaertner, S. L., Dovidio, J. F., Anastasio, P. A., Bachman, B. A., & Rust, M. C. (1993). The common ingroup identity model: Recategorization and the reduction of intergroup bias. European Review of Social Psychology, 4, 1-26. doi:10.1080/14792779343000004 Graham, J. W. (2003). Adding Missing-Data-Relevant Variables to FIML-Based Structural Equation Models. Structural Equation Modeling: A Multidisciplinary Journal, 10(1), 80- 100. https://doi.org/10.1207/S15328007SEM1001_4 Grant, S. C., & Logan, G. D. (1993). The loss of repetition priming and automaticity over time as a function of degree of initial learning. Memory & Cognition, 21, 611–618. doi: 10.3758/BF03197193 Harwood, J., Paolini, S., Joyce, N., Rubin, M., & Arroyo, A. (2011). Secondary transfer effects from imagined contact: Group similarity affects the generalization gradient. British Journal of Social Psychology, 50, 180-189. doi:10.1348/014466610X524263 Hornsey, M. (2008). Social identity theory and self-categorization theory: A historical review. Social & Personality Psychology Compass, 2, 204-222. doi:10.1111/j.17519004.2007.00066.x Imai, K., Keele, L., & Yamamoto, T. (2010). Identification, inference and sensitivity analysis for causal mediation effects. Statistical Science, 25, 51–71. doi:10.1214/10STS321 Jakobsen, J. C., Gluud, C., Wetterslev, J., & Winkel, P. (2017). When and how should multiple imputation be used for handling missing data in randomised clinical trials – a practical guide with flowcharts. BMC Medical Research Methodology, 17(1), 162. https://doi.org/10.1186/s12874-017-0442-1 39 Joyce, N., & Harwood, J. (2014). Improving Intergroup Attitudes through Televised Vicarious Intergroup Contact: Social Cognitive Processing of Ingroup and Outgroup Information. Communication Research, 41(5), 627– 643. https://doi.org/10.1177/0093650212447944 Kim, J. H. J., Lu, Q., & Stanton, A. L. (2021). Overcoming constraints of the model minority stereotype to advance Asian American health. American Psychologist, 76(4), 611– 626. https://doi.org/10.1037/amp0000799 Kline, Rex B. (2016). Principles and practice of structural equation modeling. New York :Guilford Press, Knobloch-Westerwick, S., & Crane, J. (2012). A Losing Battle: Effects of Prolonged Exposure to Thin-Ideal Images on Dieting and Body Satisfaction. Communication Research, 39(1), 79–102. https://doi.org/10.1177/0093650211400596 Long, J. A., Jeong, M. S., & Lavis, S. M. (2021). Political Comedy as a Gateway to News Use, Internal Efficacy, and Participation: A Longitudinal Mediation Analysis. Human Communication Research, 47(2), 166-191. https://doi.org/10.1093/hcr/hqaa011 Maxwell, S. E., & Cole, D. A. (2007). Bias in cross-sectional analyses of longitudinal mediation. Psychological Methods, 12(1), 23–44. https://doi.org/10.1037/1082- 989X.12.1.23 McGlone, M. S., & Aronson, J. (2006, 2006/09/01/). Stereotype threat, identity salience, and spatial reasoning. Journal of Applied Developmental Psychology, 27(5), 486-493. https://doi.org/https://doi.org/10.1016/j.appdev.2006.06.003 Nier, J. A., Gaertner, S. L., Dovidio, J. F., Banker, B. S., Ward, C. M., & Rust, M. C. (2001). Changing interracial evaluations and behavior: The effects of a common group identity. Group Processes & Intergroup Relations, 4, 299-316. doi:10.1177/1368430201004004001 Pennington, S., Beachum, C., & Holland, M. (2020, October 27). “25 best supernatural TV shows ever, ranked.” TV News. https://www.goldderby.com/gallery/best-supernatural-tv- shows-ever-ranked/supernatural-tv-ranked-wizards-of-waverly-place/ Ployhart, R. E., & Vandenberg, R. J. (2010). Longitudinal Research: The Theory, Design, and Analysis of Change. Journal of Management, 36(1), 94– 120. https://doi.org/10.1177/0149206309352110 Preacher, K. J., & Selig, J. P. (2012). Advantages of Monte Carlo confidence intervals for indirect effects. Communication Methods & Measures, 6, 77–98. doi:10.1080/19312458.2012. 679848 Roskos-Ewoldsen, D. R., Klinger, M., & Roskos-Ewoldsen, B. B. (2007). Media priming. In R. W. Preiss, B. M. Gayle, N. Burrell, M. Allen, & J. Bryant (Eds.), Mass media theories 40 and processes: Advances through meta-analysis (pp. 53–80). Mahwah, NJ: Lawrence Erlbaum Associates. Schiappa, E., Gregg, P. B., & Hewes, D. E. (2005, 2005/03/01). The Parasocial Contact Hypothesis. Communication Monographs, 72(1), 92-115. https://doi.org/10.1080/0363775052000342544 Schneider, M. (2021, May 25). “100 Most-Wathced TV Shows of 2020-21: Winners and Losers.” Variety. https://www.aol.com/100-most-watched-tv-shows-210806712.html Spears, R., Doosje, B., & Ellemers, N. (1997). Self-stereotyping in the face of threats to group status and distinctiveness: The role of group identification. Personality and social psychology bulletin, 23(5), 538-553. https://doi.org/10.1177/0146167297235009 The Walking Dead: Full Cast and Crew. (n.d.). IMDB. Retrieved June 22, 2022, from https://www.imdb.com/title/tt1520211/fullcredits Thumbore, D. (2017, October 24). “The Walking Dead” Season 8 Premiere Ratings Sink to a 5- Year Low. Collider. https://collider.com/walking-dead-season-8-premiere-ratings/ Tajfel, H., & Turner, J. C. (1979). An integrative theory of intergroup conflict. In W. G. Austin & S. Worchel (Eds.), The social psychology of intergroup relations (pp. 33– 47). Monterey, CA: Brooks/Cole. Turner, J. C., Hogg, M. A., Oakes, P. J., Reicher, S. D., & Wetherell, M. S. (1987). Rediscovering the social group: A self-categorization theory. Basil Blackwell. Turner, J. C., & Oakes, P. J. (1989). Self-categorization theory and social influence. In P. Baulus (Ed.), Psychology of group influence (2nd ed., pp. 233-275). Hillsdale, NJ: Lawrence Erlbaum. Wyer, R. S., Jr. (2004). Social comprehension and judgment: The role of situation models, narratives, and implicit theories. Mahwah, NJ: Lawrence Erlbaum Associates. Yao, S. X., Ellithorpe, M. E., Ewoldsen, D. (2022a). Supernatural genre reduces negative outgroup attitudes through strength of human identity: A replication and extension. Communication Research. International Communication Association. Yao, S. X., Ewoldsen, D. R., Ellithorpe, M. E., Van Der Heide, B., & Rhodes, N. (2022b). Gamer Girl vs. Girl Gamer: Stereotypical Gamer Traits Increase Men's Play Intention. Computers in Human Behavior, 131, 107217. https://doi.org/10.1016/j.chb.2022.107217 Zhang, Z. (2014). Monte Carlo based statistical power analysis for mediation models: methods and software. Behavior Research Methods, 46(4), 1184-1198. https://doi.org/10.3758/s13428-013-0424-0 41