Concreteness and Abstractness as Causes and Effects of Identification with Media Characters

ABSTRACT We explored new antecedents and consequences of identification with media characters that are related to concreteness or abstractness of representations. First, we examined how the portrayal of the protagonist’s behavior affects the reader’s identification with her. Participants read one of four narratives in which we manipulated the character’s usual behavior (abstract traits) and her behavior on a specific bad day (concrete states). As expected, when the character had positive traits but behaved unkindly in the negative situation, compared to the other conditions, the readers (N = 206) identified more strongly with her through the mediation of higher situational attributions, i.e. attributing the protagonist’s behavior more to the situation than to her traits. Then we examined the effect of identification on the readers’ concrete and abstract thoughts. We hypothesized and found that stronger identification with the character was related to increasingly more concrete thinking about the protagonist’s specific plans and reactions regarding the situation than abstract thinking about the protagonist’s life in general. Additionally, stronger identification was associated with more concrete thoughts about the readers themselves through the partial mediation of concrete thoughts about the character. These findings may reconcile previous inconsistent results about identification and have implications for narrative persuasion.

Communication researchers have long been interested in the psychological processes occurring while people are exposed to narratives in media.Some people establish parasocial relationships with the media characters, seeing them as friends (e.g., Derrick et al., 2008).Others identify with them, creating a merge between themselves and the media character (e.g., Cohen, 2001).This study focuses on identification and explores new antecedents and consequences of this process.The factors we believe cause identification and stem from it are both related to construal level theory (Trope & Liberman, 2010), which links the degree of closeness between an individual and a certain object to the level of relative concreteness or abstractness of mental representations.We show that presenting readers with a particular narrative about a character's positive abstract personality and concrete, situation-matching negative behavior, leads to stronger identification with the character.We also demonstrate that stronger identification contributes to more concrete thinking about the character, which relates to more concrete thinking about the self.
Construal Level Theory Trope and Liberman (2010) claim that people mentally represent their world and construe it at different levels.People may think about objects and situations at a relatively high level of construal, which is abstract, or adopt a lower level of thinking that is concrete.These levels differ in the details that are included in their representations.While abstract thinking involves integrated details that are out of context, concrete thinking involves contextualized details (Trope & Liberman, 2010).According to the theory (Trope & Liberman, 2003, 2010), construal levels are related to different types of psychological distance -temporal, spatial, hypothetical, and social -that are interrelated.Objects and events that are subjectively perceived to be closer to the self to here and to now are represented at a lower construal level (i.e., more concretely) than those that are regarded as more distant from the self, which are represented at a higher construal level (more abstractly).
Meta-analysis (Soderberg et al., 2015) suggests that the effects of psychological distance on construal level are reflected in different patterns.While many studies demonstrate a relationship between psychological distance and both abstract and concrete thinking, some studies suggest that the influence of psychological distance may selectively affect either the processing of abstract features or concrete features.
The theory maintains that the relationship between psychological distance and construal level is bidirectional (Henderson et al., 2011;Trope & Liberman, 2010).Manipulating social distance affects the level of construal (Liviatan et al., 2008), and activating a certain level of construal by encouraging abstract thinking affects perceived social distance (Stephan et al., 2010).We build on this bidirectionality and examine how incorporating concrete and abstract information in the narrative affects identification with the character (a form of social distance).We also investigate whether stronger identification promotes more concrete than abstract thinking.

Identification with Media Characters and Its Causes
When identifying with a media character, the reader or the viewer's self temporarily merges with that of the character (Cohen, Appel, et al., 2019).This experience involves seeing the events occurring in the narrative from the perspective of the character, feeling the emotions of the character, and adopting the character's point of view and motivations (Cohen & Tal-Or, 2017).Identification should not be confused with parasocial relationships, which are one-sided, friend-like bonds that the media consumer establishes with the media character (Derrick et al., 2008;Rubin & McHugh, 1987).Thus, while in identification the media consumers "lose awareness of the self as an autonomous entity and merge with the character" (Cohen, Oliver, et al., 2019, p. 79), in parasocial relationships they remain aware of themselves as separate entities that are distinct from the media character.
Research has established that people tend to identify with characters who hold attitudes similar to theirs (e.g., Cohen et al., 2015) and with those who are depicted as kind and likeable people.For example, Tal-Or and Cohen (2010) described a protagonist either as a devoted husband or a disloyal spouse who cheats on his wife.This manipulation affected identification, with participants identifying more with the faithful character.Similarly, Hoeken and Sinkeldam (2014) described a woman who suffered from a brain injury as a result of a car accident who received money from a fund designed to help people with severe medical conditions like hers receive treatment at home.The woman was described as either kind to those who helped her and thrifty in her use of the funds, or as impolite and wasteful of the money.In this study, too, the readers identified more with the kind character.
Interestingly, the effect of the likability of the characters on identification with them is not a robust finding.For example, describing a young man who suffered from cancer as likeable or unlikable did not affect identification with him (Piatok-Vaisman, 2015, as cited in Cohen & Tal-Or, 2017).Cohen and Tal-Or (2017) suggested that in cases in which we are empathic toward the character, such as when he or she is very ill, the likability manipulation does not hold.
A recent study by Tal-Or and Razpurker-Apfeld (2021a) provides a slightly different explanation for cases in which the character's likability does not lead to identification.In their study participants read a story about a woman who turned 30 but none of her friends or family remembered her birthday.The woman was depicted as behaving in a pleasant or unpleasant way.Unexpectedly, the positive depiction did not lead to more identification with the character.The authors reasoned that since behavior might be explained both as stemming from the character's personality (traits) and from the situation (state), the participants might have seen the woman who was neglected on her birthday as behaving humanly, as others would in such a situation.Indeed, when they controlled statistically for the traits attributed to the character, the authors found that people identified more with the rude and unpleasant character.A major goal of our study is to examine this hypothesis and explore whether identification is affected differently when the same behavior is perceived as stemming from the character's traits or state.

Traits and States as Predictors of Identification
Attribution theorists have long asserted that human behavior can be attributed either to a situation that puts people in a specific state or to one's personality or traits that explain individual differences in behavior when experiencing the same state.States are concrete as they can be easily detected and observed.In contrast, traits are abstract.They can only be deduced from behavior in different times and circumstances.Furthermore, states are momentary, whereas traits are considered stable and long-lasting (Fridhandler, 1986).
According to construal level theory, trait attributions are abstract personality dispositions that are prototypical and decontextualized.They are readily attributed to other people, who by definition are more psychologically distant from oneself.In contrast, situational attributions are concrete.They focus on the unique context that guides behavior, and are spontaneously attributed to oneself -the closest social distance (Trope & Liberman, 2010).Given the bidirectionality in construal level theory, situational attribution may reduce the distance from the self and increase identification with media characters.However, in line with studies documenting the effect of the valence of personality on identification (e.g., Tal-Or & Cohen, 2010), we predict that situational attribution will increase identification with media characters only when their positive personality is combined with realistic negative behavior.
We argue that when media characters' behaviors seem to stem from their personality, people will tend to identify with the positively depicted character.However, when a character faces a difficult situation, his or her negative behavior might be attributed to the situation, not to his or her personality.This character can be perceived as reacting to the situation in a typical and realistic way.We hypothesize that people will identify more with a character with a positive personality who behaves negatively in situations where the reader or viewer could see themselves behaving similarly under such circumstances.
This assumption is in line with the classic "pratfall" effect discovered by Aronson et al. (1966), according to which people tend to like successful people more when they are not perfect and are seen as human.This is especially true for people with an average level of self-esteem who feel more similar to the not-so-perfect successful person (Helmreich et al., 1970).Researchers such as Hoeken et al. (2016) have documented that perceived similarity between media consumers and media characters enhances identification.Thus, it would be much easier to identify with characters who behave humanly, like the media consumers.Moreover, previous research has demonstrated that people tend to identify more with characters in a narrative that media consumers regard as realistic, meaning, it depicts a situation similar to one they have experienced and the characters' behaviors are similar to those of people in the real world of the media consumers (Cho et al., 2014).Taken together, we expect people to identify more with characters who behave humanly and realistically, as most people would behave in response to a negative situation.More specifically, we hypothesized that: H1: Reading about a character who usually behaves positively but behaves unkindly in a specific negative situation will lead to more situational than trait behavior attribution than reading about a character whose behavior is always positive and characters whose behavior is always or usually negative.
H2: Reading about a positive character who behaves unkindly in a specific situation, compared to the other narratives, will be related to stronger identification with that character through the mediation of more situational than trait behavior attribution.
Thus, we predict that in certain cases more concrete descriptions (states) relative to abstract descriptions (traits) play a role in the degree to which people identify with a particular media character.In addition, we believe that identification affects the level of concreteness or abstractness of the thoughts about the protagonist and the intended behavior of the media consumer.

The Effects of Identification on Attitudes and Construal Levels of Thought
The interest of scholars in the process of identification with media characters stems mainly from its documented effects on media consumers.For example, identification with media characters predicts enjoyment from watching the narrative (e.g., Igartua, 2010).Identification also influences perceptions about the self (Sestir & Green, 2010).Most importantly, identification plays a role in affecting the attitudes of media consumers to accord with those expressed by the media character (e.g., de Graaf et al., 2012).There is also a significant positive correlation between the level of identification with media characters and behavioral intentions (e.g., Kresovich & Noar, 2020).We argue that the level of identification can affect the types of thoughts the readers have about the character, and more specifically, their level of concreteness or abstractness.We also examine the implications for the readers' intended behaviors.
As mentioned above, the construal level theory (Trope & Liberman, 2010) asserts that the degree of closeness between an individual and a certain object is related to different construals of mental representations.For example, Young (2015) reported that hearing a persuasive health message from someone people regard as similar to them prompted them to focus on the feasibility of the health behavior which is considered concrete because it involves specific aspects that can hinder or foster the behavior.
When media consumers merge with the media character, they are presumably very close in terms of psychological-social distance.Thus, relying on construal level theory and findings (Soderberg et al., 2015;Trope & Liberman, 2010) we predict that stronger identification will be associated with relatively more concrete than abstract thinking, manifested in having stronger contextualized thoughts about the character such as how the character is expected to behave in reaction to the events in the story, compared to decontextualized thoughts such as the changes the character has to make in life.
While in identification people feel as if they are the media characters, in parasocial relationships they see them as separate entities (Moyer-Gusé, 2015), and thus relatively remote.According to construal level theory, psychologicalsocial distances are experienced with reference to the self.Therefore, less divergence from the self is associated with more concrete processing (Liberman & Trope, 2014;Trope & Liberman, 2010).The range of the psychological distance is greater in identification (from character-self mergence to character-self divergence) than in parasocial relationships (from character-self close relationships between two entities to character-self distant relationships) (see Figure 1).As a result, the expected differences in construal level for readers with different levels of identification with the character may not be evident when a narrowed spectrum of psychological distance is examined, as in the case of parasocial relationships with the character.We therefore hypothesized that: H3: Identification will have a moderating effect on construal level, whereby stronger identification with the protagonist will relate to lower construal levels of thoughts about the protagonist (more concrete than abstract thinking).In contrast, parasocial relationships will not have a moderating effect on construal level of thoughts about the protagonist.
If confirmed, and provided that in identification the media consumer merges with the media character (Cohen, Oliver, et al., 2019), we further expect that: H4: Identification with the protagonist will be associated with stronger concrete thoughts about the self through the mediation of stronger concrete thoughts about the protagonist.

Participants and Data Collection
Sample size planning was based on two types of calculations.First, an a priori power analysis using G*Power (Faul et al., 2009) indicated that a sample of 175 participants could reveal close to medium size effects (f = .213,based on previous work, Tal-Or & Razpurker-Apfeld, 2021a) with a power of .80 for a design including four groups and controlling for covariates.Second, we followed the guidelines presented in simulation studies on the statistical power needed to test for mediation (Fritz & MacKinnon, 2007).Accordingly, if the "a" and "b" paths in the model were of small-to-medium sizes (β = .26),a sample size of 148 or larger would be required to detect a significant mediation effect using the bias-corrected bootstrap method with a statistical power of .80.
Thus, we distributed a link to our study to 232 potential participants through Panel4All, an Israeli online crowdsourcing platform with tens of thousands of members who represent various demographic segments of Israeli society (https://www.panel4all.co.il).The inclusion criteria for the study included native Hebrew speakers who were 20 to 40 years old, 50% male and 50% female.Six participants who did not agree to the consent form were immediately removed and not given any of the study's materials or questions.Participants who failed the multiple-choice questions about the story were excluded from the study.The other participants were rewarded for completing the study in accordance with Panel4All's policy.The final sample included 206 participants, 105 women and 101 men, ages 20-40 (M = 30.29,SD = 5.70).All of our procedures met the ethical standards of our institutions' research committees for experiments involving human subjects (approval number 17-2021-a from Zefat Academic College and approval number 259/21 from the University of Haifa).

Design and Procedure
We manipulated two independent variables.The first referred to the usual behavior of the protagonist which implies about her personality (positive versus negative).The second described the protagonist's behavior on aspecific day on which her family and friends forgot her 30 th birthday (kind versus unkind behavior).The participants were randomly assigned to one of four experimental conditions in the 2 (usual behavior: positive/negative) X 2 (behavior today: kind/unkind) design (see Figure 2).First, we examined the effect of these two variables on identification (H2).Then, we examined identification as a moderator (H3) and as an independent variable (H4).
Participants read one of four versions of a story depending on the experimental condition to which they were assigned.Then, they filled out a questionnaire that included all of the measures in the study, as detailed below.The story and the questionnaires were presented in Hebrew.All questionnaires were prepared and administered using Qualtrics software (Qualtrics ©, http://www.qualtrics.com/uk/,2019).After answering the main questionnaires, which included the study's measures, the participants were asked whether they thought the protagonist was nice on the specific day depicted.We also asked them two simple multiple-choice questions to verify their basic understanding of the story's content and excluded those who failed to answer correctly. 1We also asked the participants whether they suspected anything during the experiment and what they thought was the purpose of the study.Finally, participants provided demographic details about their gender, age and mother tongue. 2 Upon completion, the participants received a written debriefing.

Materials
We prepared four variations of a story used in a previous study (Tal-Or & Razpurker-Apfeld, 2021a).Each story was about 1,200 words long.Rotem was a female protagonist who was turning 30 on the specific day described.The story described Rotem's general traits and her usual behaviors and also described her thoughts and behaviors on her birthday itself.It depicted Rotem's day at home and at work, and focused on her ongoing disappointment, as it seemed that everyone had forgotten her birthday.
Four narratives were prepared based on a combination of the two manipulated variables (see Figure 2).We attempted to portray the intended personality and situational behavior by using rich, detailed descriptions to present the main protagonist's traits, thoughts and behaviors.We also included other characters in the narrative to comment on Rotem's behavior on the specific day, and whether it fit her general behavior.Sample sentences from the stories that distinguish the four variations and present Rotem as intended are provided in the supplementary material associated with this paper.

Perceptions about the general traits of the protagonist
This measure was adapted from previous studies (Tal-Or, 2019; Tal-Or & Razpurker-Apfeld, 2021a).Participants indicated the extent to which Rotem's character had the following traits -good-hearted, rude, anxious, pleasant, aggressive and kind -on scales ranging from 1 (does not characterize her at all) to 7 (characterizes her a lot).After reverse coding three items we calculated an index based on the mean of Rotem's perceived personality traits (M = 4.43, SD = 1.82,Cronbach's α= .96).Higher values of this index indicated that readers perceived Rotem as having more positive traits (being generally kinder).

Identification with the protagonist
We adapted these items from Cohen (2001) and asked the participants to indicate the extent of their agreement with seven statements regarding Rotem, such as: "While reading, I felt like Rotem felt," "While reading, I could really get into Rotem's head," "At some moments while reading, I felt I knew exactly what Rotem was going through."Responses were made on 1 ("strongly disagree") to 7 ("strongly agree") scales.We calculated a mean identification index (M = 5.00, SD = 1.38,Cronbach's α= .91),for which higher values indicated stronger identification with Rotem.

Parasocial relationships with the protagonist
We based this measure on two relevant factors from Tukachinsky's (2010) multiple parasocial relationships scale.Specifically, we adapted the friendship communication factor and the friendship support factor.Statements such as "If Rotem were a real person, I could have disclosed a great deal of things about myself to Rotem," "If Rotem were a real person, I think Rotem could be a friend of mine," and "If Rotem were a real person, I could trust her completely" were presented.We asked the participants to indicate the extent to which they agreed with 13 statements on scales from 1 ("strongly disagree") to 7 ("strongly agree").We calculated a mean parasocial relationships index (M = 4.35, SD = 1.41,Cronbach's α= .96).Higher values indicated stronger parasocial relationships.

Transportation
We wanted to make sure that the stories are engaging enough and that they do not differ one from another in that respect.Thus, we measured the extent to which the readers were absorbed in the narrative using 11 items from Green and Brock's transportation scale (Green & Brock, 2000).We asked the participants to indicate the extent to which they agreed with statements such as: "I was mentally involved in the story while reading it" and "The story affected me emotionally" on 1-7 scales (1 = "strongly disagree," 7 = "strongly agree").After dropping one item ("While I was reading the story, activity going on in the room around me was on my mind") that reduced the reliability below the criterion of .65,we calculated a mean score which had a satisfactory reliability (M = 4.49, SD = .87,Cronbach's α = .75).

Construal level thinking with regard to the protagonist
Based on the distinctions made in previous research between concrete and abstract construals (see Liberman & Trope, 2014;Moran et al., 2021), we developed three statements that were intended to tap concrete, low-level construals, and three statements that were intended to tap abstract, highlevel construals with regard to the protagonist, and presented them in the same block of questions.Concrete construals included the following statements: "While reading, I thought what Rotem should do on her birthday to feel better," "While reading, I thought what was going to happen to Rotem in the next few hours" and "While reading, I thought how Rotem should react if by the end of the day no one remembered and celebrated her birthday."Abstract construals included the following statements: "While reading, I thought of the changes Rotem should make in her life to be in a better position," "While reading, I thought of the future awaiting Rotem" and "While reading, I thought about the long term consequences of the described situation for Rotem."Participants had to indicate the extent to which they agreed with the statements on scales from 1 ("strongly disagree") to 7 ("strongly agree").An initial principal component analysis (PCA) with Varimax rotation and Kaiser normalization demonstrated that out of the six items, five exclusively represented either the concrete or abstract factors.The first item that was intended to measure concrete thinking was common in both factors, and was therefore removed from the analysis.The other items (loadings > .75)captured the concrete thinking factor (42% explained variance) and the abstract thinking factor (31% explained variance).Then, for each participant we calculated a mean "concrete thinking about Rotem" score (M = 5.05, SD = 1.57,Spearman-Brown coefficient = .70),and a mean "abstract thinking about Rotem" score (M = 4.33, SD = 1.60,Cronbach's α= .82),with higher values representing more concrete or more abstract thinking, respectively.

Concrete thinking with regard to the self
Following a similar procedure described for the previous measure, we developed three statements that were intended to tap concrete, low-level construals with regard to the self, and how the readers would specifically react if they were in a situation similar to that of the character: "While reading, I thought what I would tell my family and friends had they not congratulated me on my birthday," "While reading, I thought what I would do in a situation similar to Rotem's" and "While reading, I thought how I would feel if my birthday were forgotten."For each participant we calculated a mean "concrete thinking about the self" score (M = 4.75, SD = 1.63,Cronbach'sα = .83).Higher values represented more concrete thinking regarding the self. 3

State-trait attribution of the protagonist's behavior
We presented a block of statements designed to tap situational attributions and trait attributions for Rotem's behavior.Situational attributions included the following statements: "Rotem's behavior is appropriate to the situation in which she was," "Being in Rotem's situation, I would behave so too," "It seems that Rotem's behavior on the day depicted in the story stems from the situation she was in" and "Rotem's behavior on the day depicted in the story was justified."Trait attributions included the following statements: "It seems that Rotem's behavior on the day depicted in the story characterizes her usual behavior" and "It seems that Rotem's behavior on the day depicted in the story stems from her personality."Responses were made on scales from 1 ("strongly disagree") to 7 ("strongly agree").PCA with Varimax rotation and Kaiser normalization demonstrated that all six items represented the situational and trait attribution factors as intended.The items (loadings > .66)captured the situational attribution factor (46% explained variance) and the trait attribution factor (29% explained variance).Then, for each participant, we calculated a mean "situational attribution" score (M = 4.73, SD = 1.43,Cronbach's α= .84),and a mean "trait attribution" score (M = 4.31, SD = 1.77,Spearman-Brown coefficient = .81).These two scores had a negligible correlation (r = .09).Finally, for each participant we calculated a difference index of attribution by subtracting the trait attribution score from the situational attribution score. 4Higher values represented relatively more situational attribution than trait attribution (M = .42,SD = 2.18, ranging from −5.50 to + 6.00).The estimated reliability of the difference score was good (r DD = .81).The situational attribution component positively correlated with the difference score (r = .59),and the trait attribution component negatively correlated with the difference score (r = −.76).

Preliminary Checks
None of the participants guessed the purpose of the study or suspected anything unusual.Initially, we examined the general understanding of the narratives and the transportation into them and tried to confirm that there were no differences among the four experimental conditions with regard to these aspects.Similar to previous studies (Cohen et al., 2015;Green & Brock, 2000), the written narrative produced a relatively high transportation score (M = 4.49, SD = .87).Indeed, the participants who agreed to participate in this study and were allocated to the four experimental story conditions did not differ significantly in their transportation into the narratives, F < 1.
Furthermore, most of the participants were correct in their responses to both of the multiple-choice questions about the story (91.2%).Participants in the four story conditions did not differ in their understanding of the details concerning a question about the character's special occasion, χ 2 9 ð Þ = 6.57, p = .682,and a question about the character's activity on the birthday morning, χ 2 6 ð Þ = 4.11, p = .661.Participants who did not answer both of these questions correctly were removed from the subsequent analyses.Answering both questions correctly (or not) did not depend upon the story's condition, χ 2 3 ð Þ = 6.30, p = .098. 5  Next, we checked if the narratives were understood as intended.We conducted a t-test for independent samples to examine whether the readers of the different stories perceived the difference in the protagonist's behavior on the specific day.The story condition depicting the behavior on the specific day (kind today vs. unkind today) was the independent variable and the perceived kindness on the depicted day was the dependent variable.A significant effect was found, t(204) = −10.969,p < .001,Cohen's d = 1.53.When Rotem was described as kind on the specific day, she was perceived as kinder, M = 5.42, SD = 1.60, compared to when she was described as unkind, M = 2.85, SD = 1.76.
Another t-test was conducted, in which Rotem's personality (usually behaves positively or negatively) was the independent variable and the perceived general traits of the protagonist was the dependent variable.A significant effect was found, t(204) = −10.877,p < .001,Cohen's d = 1.53.When Rotem's personality was described positively, she was indeed perceived more positively, M = 5.59, SD = 1.30, compared to when she was described negatively, M = 3.38, SD = 1.58.Thus, Rotem's traits were perceived as intended.These two manipulation checks confirmed that the readers in the different story conditions understood Rotem's personality traits and her behavior in the specific situation on her birthday.
Given that the main character in the stories was female and the readers were female or male, we examined their distribution across the experimental conditions.The distribution of men and women in the four experimental groups did not differ, χ 2 3 ð Þ = 2.94, p = .402.Furthermore, we examined whether the identification with the main character differed based on gender.We found an almost significant effect, t(204) = −1.94,p = .054,Cohen's d = .27.Women identified more with the female protagonist, M = 5.18, SD = 1.34, than men, M = 4.81, SD = 1.41.Thus, we decided to control for gender in the following analyses when identification was the dependent variable. 6

Main Analyses
An initial general 2 × 2 analysis of covariance (ANCOVA) examined the effects of the manipulations of Rotem's usual behavior and her behavior on the specific day on identification with her while controlling for the participants' gender.The analysis showed no significant main effect of the manipulated behavior on the specific day and no interaction, F s <1.Only the main effect of the manipulated usual behavior had a significant effect on identification, F(1,201) = 8.257, p = .004,MSE = 1.832, η 2 p = .039.Participants identified more strongly with Rotem when she was depicted as usually positive (adjusted M = 5.28, SE = .14)than when she was depicted as usually negative (adjusted M = 4.73, SE = .13). 7 We turn now to the focal research questions.The first two hypotheses contrasted the specific story condition of "positive personality and situational unkindness" with the other three conditions in which Rotem was always positive, always negative, or usually negative but kind in the negative situation.We examined H1 and H2 using a mediation model with the SPSS PROCESS macro (Hayes, 2017; Model 4 with 5,000 bootstrapped resamples) in which Rotem's behavior ("positive personality and situational unkindness" vs. the other conditions) was the independent variable, identification with her was the dependent variable, and the difference attribution index was the mediator, while controlling for gender.
The results indicated that reading about the character who was usually positive but behaved unkindly in a negative situation led to more situational than trait attributions to her behavior, compared to the other conditions, b = .98,SE = .37,p = .008,in line with H1.Stronger situational than trait attributions were related to significantly stronger identification with Rotem, b = .21,SE = .04,p < .001.The total effect of Rotem's behavior on identification was not significant, b = .24,SE = .23,p = .306,95% CI [−.407, .480],and neither was the direct effect, b = .04,SE = .23,p = .877,95% CI [−.413, .483],but there was a significant indirect effect, IE = .20,SE = .09,95% CI [.051, .405],thus confirming H2 (see Figure 3). 8The description of Rotem's behavior in the specific story (positive personality but unkind in the situation) compared to the other stories was related to stronger identification through the mediation of higher situational than trait attributions.
H3 predicted that stronger identification with the protagonist, but not stronger parasocial relationships, would be related to lower construal levels of thoughts about the protagonist.To examine H3 we employed Unstandardized regression coefficients are presented together with standardized coefficients (in brackets).The direct effect and the total effect (in square brackets) were non-significant but the indirect effect (depicted with a gray dotted line) was significant.CI stands for 95% confidence intervals.***p< .001.
a repeated measures additive moderation model using the SPSS MEMORE 2.1 macro (Montoya, 2019; Model 2 with 5,000 bootstrapped resamples).This approach allowed us to examine the moderation effects of continuous variables on repeated measures.In the model, identification and parasocial relationships were treated as between-subjects moderators, while construal levels regarding the media character were considered within-subjects dependent variables.Difference dependent scores were derived from the analysis.A positive value of the difference score indicated a higher tendency toward concrete thinking (lower construal level) compared to abstract thinking.Additionally, we included the covariates of the manipulated conditions (Rotem's behavior on the specific negative situation and her usual behavior) as additional moderators.
As expected, the analysis revealed a significant moderation effect of identification on construal level, b = .21,SE = .10,p = .043,95% CI [.007, .414],indicating that stronger identification with the character was associated with increasingly lower construal levels (more concrete thinking and less abstract thinking).In contrast, parasocial relationships did not exhibit a moderating effect on construal level, b = −.07,SE = .10,p = .506,95% CI [−.273, .135].This suggests that the readers' construals remained stable irrespective of their parasocial relationships with the character.
To examine H4 about the mediated effect of identification on the level of concreteness of thoughts about the self, we used a mediation model (SPSS PROCESS, Model 4) in which the independent variable was identification with Rotem, the dependent variable was the reader's concrete thoughts about the self, and the mediator was the reader's concrete thoughts about Rotem.We controlled for both of the manipulated conditions.The results indicated that higher levels of identification were correlated with more concrete thoughts about Rotem, b = .58,SE = .07,p < .0001,and that more concrete thoughts about Rotem were correlated with more concrete thoughts about the self, b = .57,SE = .06,p < .0001.The total effect of identification on concrete thoughts about the self was significant, b = .60,SE = .07,p < .0001,95% CI [.460, .748],and so was the direct effect, b = .27,SE = .07,p = .0002,95% CI [.134, .413].Furthermore, in line with our expectations, a significant indirect effect was found: stronger identification was related to more concrete thoughts and behavioral intentions about the self through the mediation of concrete thoughts about the character, IE = .33,SE = .06,95% CI [.212, .458]. Figure 4 presents this partial mediation.

The Contribution of Abstract and Concrete Representations to Identification
We examined the relationship between the concreteness and abstractness of representations of media characters and level of identification with them.We showed that although people generally identify with a protagonist with a positive personality (e.g., Tal-Or & Cohen, 2010), they might also take into consideration states that prompt the protagonist to behave unkindly.In the context of this study, being exposed to the particular narrative and its characters, people tended to identify with a good protagonist who behaved negatively in a bad situation, attributing her behavior to the situation.
Construal level theory asserts that situational attribution is related to psychological closeness.However, our findings show that attribution alone cannot explain identification, as the protagonist's personality also matters.Based on previous research we know that there are other factors besides abstract and concrete representations that predict identification with media characters.For example, people tend to identify with characters who are presented as having a more positive personality (e.g., Hoeken & Sinkeldam, 2014).Therefore, we placed a positive character in a negative situation that prompted unkind A mediation model demonstrating the predicted effect of the reader's identification with the character on concrete thoughts about the self through concrete thoughts about the character (H4).Note.A mediation model (Process Model 4) was conducted to test H4 (N = 206), while controlling for the two manipulated variables.Unstandardized regression coefficients are presented together with standardized coefficients (in brackets).The direct effect was significant and so was the indirect effect (depicted with a gray dotted line).The significant total effect is also presented (in square brackets).CI stands for 95% confidence intervals.***p < .001.
behavior that was natural and realistic for this specific situation, so situational attribution increased identification.If, however, a character with a negative personality were put in a positive situation that necessitated kind behavior and behaved kindly in that situation, we would not expect situational attribution to lead to more identification, although construal level theory would predict so, since people do not tend to identify with characters with a negative personality.
We show that identification depends on the attribution attached to the behavior.If the bad behavior seems to stem from the character's personality, media consumers will be less likely to identify with the character.However, if the bad behavior is perceived as resulting from the situation, people will be more likely to identify with the media character.This finding explains the inconsistent results with regard to the effect of the character's virtue on identification with him or her (Cohen & Tal-Or, 2017).In Tal-Or and Cohen (2010) the narrative did not contain any situational explanation for the bad behavior of the protagonist.Thus, in that case, the manipulation of the character's virtue worked.In Tal-Or and Razpurker-Apfeld (2021a) the protagonist experienced a difficult situation when she was neglected on her birthday.Thus, her negative behavior, including acting nervously and impolitely, did not reduce identification with her.This was also the case in Piatok-Vaisman (2015, cited in Cohen & Tal-Or, 2017) in which people did not identify less with a young cancer patient who behaved impolitely.Here too, the rude behavior might have been perceived as stemming from his situation, and thus did not reduce the identification with him.The manipulation of the media character's virtue did work in Hoeken and Sinkeldam (2014) in which the protagonist whose behavior was positively or negatively depicted was a very ill woman.However, a large part of the manipulation regarding the woman's behavior revolved around her use of the financial budget she received from the government.In one scenario she was described as an exploiter who did not always pay her friend who helped her and kept the rest of the money for herself.In the contrasting scenario, she was described as a decent person who paid her friend, did not use all of the funding and returned the rest of it to the government.Arguably, the character's specific kind of negative behavior could not be explained by the woman's situation and thus seemed more indicative of her inner personality.Future research might manipulate the type of negative behavior to match or not match a particular negative situation and examine whether indeed the reduced negative effect of the character's behavior on identification with her occurs only in a matching condition.
Interestingly, the condition in which the character usually behaves positively but behaved unkindly in a specific negative situation revealed more identification with the character than the other conditions only through the mediation of attributing the character's behavior to the situation.There were no direct and total effects.This might be the result of other mediators that attenuate the effect.Future research should examine the possible effects of other mediators using the particular story of the current study and also generalize these effects with other stories.

The Influence of Identification on Construal Level of Thinking
Thus, in the first part of the study we demonstrated the effect of construal level on identification with the media character.We then explored how identification influences the construal level of thoughts.We found that readers who identified more strongly with the character and maintained more psychological proximity with her used more concrete than abstract thinking about her.This difference in construal level became less prominent as readers identified less with the character.Construal level theory predicts that reducing psychological distance will lead to relatively more concrete than abstract thinking (Soderberg et al., 2015;Trope & Liberman, 2010).Our findings support the theory by using identification as a measure of psychological distance.The specific pattern of results is similar to other findings that support the theory showing that psychologically closer events are perceived in more concrete than abstract terms and that this difference was larger than when psychologically distant events were perceived (e.g., Ledgerwood et al., 2010 -Study 1).
Interestingly, in contrast to identification, parasocial relationships, where readers and the protagonist are considered as separate entities (Moyer-Gusé, 2015), did not show any changes in construal levels of thinking associated with the strength of those relationships.This finding supports other work suggesting that identification and parasocial relationships are separate constructs with different consequences (e.g., Tal-Or & Razpurker-Apfeld, 2021b).
More importantly, however, we showed that more identification with the character contributes to more concrete thoughts about what one would do in the character's shoes.This connection between identification and thinking about one's own possible concrete behavior worked directly and also indirectly through thinking about the character's possible concrete behavior.Since in identification the reader and the character merge (Cohen, Appel, et al., 2019), identification seems to affect thinking about the character and thinking about the self interchangeably.Our findings indicate the importance of strong identification for concrete thinking about one's own behavior, which has clear implications for narrative persuasion.
We asked our participants to report their thoughts about the hypothetical behavior they would have engaged in if faced with the situation depicted in the story.This measure is different from the common measures of behavioral intentions that ask participants to report their intentions to perform specific behaviors (e.g., Greaves et al., 2013).Future research might study the effect of a narrative with a clear persuasive message on behavioral intentions.Extending our findings beyond the chosen narrative that focused on a specific character disappointed at her birthday, we carefully suggest that if the narrative intends to influence a particular, concrete behavior, such as not drinking and driving or using seat belts, then enhancing the audience's identification with a media character would be recommended.However, if the aim is to make people think more globally and abstractly, then identification will not play such a dominant role.This direction deserves further exploration with new narratives.

Limitations and Future Research
Like any other research, ours is not free of limitations.First, we based our research on one specific narrative and a main character, similar to that used in a previous study (Tal-Or & Razpurker-Apfeld, 2021a).This may limit our contribution to cases in which the protagonist is a young female, her "bad" behavior is externalized and reflected in grumpiness, and the event is a forgotten birthday.The replicability of our findings should be tested with different stories and different characters, which might include other variables that influence identification in other ways.For example, what would happen if the protagonist behaved badly in a way that was in line with the situation, but by doing so harmed other people?Or what happens if the protagonist who behaves badly is part of an unfavorable outgroup?However, while our study was based on a specific narrative, our hypotheses were based on the findings of previous studies that used other narratives (e.g., Hoeken & Sinkeldam, 2014;Tal-Or & Cohen, 2010).The attribution explanation we suggested seems to be consistent with the findings of these other studies.Still, incremental evidence from multiple studies aiming to explore the role of traits and states of protagonists in various stories on the reader's identification will be beneficial.
Second, our measure of the dependent variable of level of abstractness and concreteness was developed specifically for this study.There is no widely used scale that measures this construct and the various studies that examined construal level measured it in various ways (e.g., Fujita et al., 2006;Liviatan et al., 2008;Moran et al., 2021;Nussinson et al., 2021).Thus, the construct validity of the scale should be examined using other measures.For example, it would have been interesting to examine actual behavior instead of behavioral intentions.
Lastly, while the first part of the manuscript about the effects on attribution and identification is experimental in nature, the second part related to the effects of identification is correlational.Thus, we should be cautious in drawing conclusions about the causal effects of identification.Future research might examine the effect of identification on the construal level of thoughts experimentally by manipulating identification levels.

Conclusions
Notwithstanding its limitations, we believe our study adds to our understanding about the antecedents and consequences of identifying with media characters.In line with construal level theory, we demonstrated that lower construal levels in attributing the behavior of a character possessing a positive personality leads to enhanced identification, and that identification relates to lower construal levels of thoughts regarding the character.More specifically, we demonstrated that at least for contexts similar to that set by our narrative, people take into account not only the valence of the protagonist's behavior but also its suitability to the depicted situation.It enables them to view the behavior as more concrete and not indicative of an abstract trait, which has implications for identification.
We also showed how identification affects the level of concreteness and abstractness of thoughts, and how concrete thinking about the character's specific actions in the narrative relates to the reader's thinking about his or her own concrete actions in a similar situation.We look forward to future studies that continue to explore the bidirectional relationship between construal level and engagement with media characters.

Notes
1.The questionnaire also included a measure of enjoyment of the movie and questions regarding the perceived sociability of the protagonist.These questions were outside the scope of the current study, and thus were not analyzed.2. The data underlying this article will be shared on request.3. We also measured abstract thinking with regard to the self but we do not elaborate about it.We intended to use difference scores for concrete vs. abstract thoughts for the mediation in H4.As the difference score with regard to the character was not reliable, we did not use any difference score in H4, and the mediation focused only on measures of concrete thinking.4. Using difference scores has been criticized in various research fields as being less reliable than their components (e.g., Cafri et al., 2010).Others, however, have mentioned the common use of difference scores in making a noteworthy contribution to understanding phenomena in cognitive psychology and social psychology, if applied with caution based on the case in hand (e.g., Trafimow, 2015).We believe that using a difference "situational minus trait" score as a dependent variable (H1) and a mediator (H2) in the current study does reflect meaningful theoretical information.The score is acceptable if the component scores themselves are reliable and do not have high correlations with each other (Tisak & Smith, 1994).Moreover, as recommended (Furr, 2011), we present descriptive statistics and information about its psychometric quality in addition to information about the separate components.The estimated reliability of the difference score (r DD ) was calculated according to an adequate equation (Equation 7.6, Furr, 2011).
5. A total of 20 participants were dropped due to failure to respond correctly to both questions.Participant loss occurred in all four story conditions: negative personality and situational unkindness (n = 2), positive personality and situational kindness (n = 5), positive personality and situational unkindness (n = 9), negative personality and situational kindness (n = 4).6.To determine the need for controlling for gender when identification served as an independent variable, we conducted additional analyses to assess the impact of gender on other variables.The results indicated that gender had no significant effect on the construal level regarding the media character (concrete vs. abstract), F < 1.Furthermore, we found that gender did not significantly impact the readers' concrete thoughts about Rotem or their concrete thoughts about the self, Fs <1.Based on these findings, we decided to control for gender when examining H1 and H2, while not including it as a control variable in H3 and H4. 7. Running the analysis without gender as a covariate yielded the same pattern.Specifically, there was a significant main effect of the manipulated usual behavior, F(1,202) = 7.442, p = .007,MSE = 1.861, η 2 p = .036,while the other main effect and the interaction did not reach significance level.8. Running the analysis again without controlling for gender did not impact our findings.
Specifically, reading about the character's situational unkindness led to more situational than trait attributions to her behavior, compared to the other conditions, b = .97,SE = .36,p = .008,supporting H1, and the character's situational unkindness was related to stronger identification through the mediation of more situational than trait attribution, IE = .20,SE = .09,95% CI [.045, .397],supporting H2.

Figure 1 .
Figure 1.A conceptual sketch of the varying ranges of psychological-social distance for identification and parasocial relationships.

Figure 2 .
Figure 2. The presentation of the protagonist in the different experimental conditions.

Figure 3 .
Figure 3.A mediation model demonstrating the predicted effect of the character's behavior on the reader's identification through more situational behavior attribution (H2).Note.A mediation model (Process Model 4) was conducted to test H2 (N = 206), while controlling for gender.Unstandardized regression coefficients are presented together with standardized coefficients (in brackets).The direct effect and the total effect (in square brackets) were non-significant but the indirect effect (depicted with a gray dotted line) was significant.CI stands for 95% confidence intervals.***p< .001.

Figure 4 .
Figure4.A mediation model demonstrating the predicted effect of the reader's identification with the character on concrete thoughts about the self through concrete thoughts about the character (H4).Note.A mediation model (Process Model 4) was conducted to test H4 (N = 206), while controlling for the two manipulated variables.Unstandardized regression coefficients are presented together with standardized coefficients (in brackets).The direct effect was significant and so was the indirect effect (depicted with a gray dotted line).The significant total effect is also presented (in square brackets).CI stands for 95% confidence intervals.***p < .001.