Character identification is predicted by narrative transportation, immersive tendencies, and interactivity

The question of what contributes to users’ identification with media characters remains an open issue in research. Apart from media interactivity, user characteristics like gender, age, immersive tendencies, and factual transportation into the narrative are promising factors. Yet, research is still in its infancy, and the usage of different media limits cross-study comparability. Hence, the current study set out to examine predictors of character identification using a text-based fiction with purported interactivity which was inspired by interactive fiction (IF) games. In an online experiment, 228 participants aged 15–65 years were randomly assigned to either an active condition where they could choose from different options, or a passive condition where they only read the story. Additionally, participants filled out questionnaires assessing immersive tendencies, level of identification, and narrative transportation. A multiple linear regression model tested for predictors of character identification. Apart from age and gender which remained non-significant, interactivity, immersive tendencies, and transportation into narrative significantly predicted identification with the IF’s main character. The current findings support theoretical models on media interactivity and identification, yet several open issues such as the role of media content (engaging vs. mundane) and character features (e.g., similarity with user) remain to be answered.


Introduction
Identifying with a fictional character lies at the core of many media experiences, be it reading a novel, watching a movie, or playing a video game (Cohen, 2001). Past studies have repeatedly linked identification with media characters to increased media enjoyment (Birk et al., 2016;Busselle & Bilandzic, 2009;Hefner et al., 2007;Trepte & Reinecke, 2010), more positive affect and empathy (Bachen et al., 2012;Birk et al., 2016), a facilitated experience of flow (Soutter & Hitchens, 2016), greater interest in learning (Bachen et al., 2016), and enhanced persuasiveness of media content (Green & Brock, 2000;Moyer-Gusé et al., 2011;Slater & Rouner, 2002). As such, character identification is thought to shape the success of applications that seek to induce behavior or attitude changes in the context of entertainment-education (see Bachen et al., 2016;Green & Jenkins, 2014;Slater & Rouner, 2002) or health care (see Steinemann et al., 2017). Apart from these beneficial effects, however, identification has also been shown to be associated with negative media consequences such as addictive 1 3 behaviors (Smahel et al., 2008) and aggression (Bachen et al., 2012;Konijn et al., 2007).
While prior conceptualizations of identification (see Moyer-Gusé, 2008, for an overview) have focused on how readers or viewers felt about the character in terms of sympathy or similarity, current definitions put their emphasis on how readers, viewers or gamers feel with the character (Cohen, 2001). Hence, identification is construed as "an imaginative process through which an audience member assumes the identity, goals, and perspective of a character" (Cohen, 2001, p. 261). As such, a person reading a book, watching a movie, or playing a video game will experience and interpret the text, film or game as if she or he was on the inside and as if the described events were happening to her or him. In other words, identification may be described as "a temporary alteration of media users' self-perception" (Hefner et al., 2007, p. 40) which, on the one hand, involves a transitory loss of self-awareness and, on the other hand, encompasses adopting the character's point of view (cognitive empathy) and feeling what the character is feeling (emotional empathy) (Cohen, 2001).

Predictors of Character Identification
Although a considerable body of literature has explored the effects of character identification on media consumers, findings are still inconsistent with regards to the question of what contributes to identifying with a fictional character. Identification is hypothesized to be influenced by a range of factors, including not only different features of the medium (e.g., level of interactivity, narrative content) but also specific attributes of the character (e.g., appearance, customization options), and a range of user characteristics (e.g., gender, age, engagement with content) (see Cohen, 2001).
Among these, interactivity has received considerable attention in the past, and several theoretical approaches have incorporated interactivity as a causal factor. For instance, the Conceptual Model of Interactivity Effects (Green & Jenkins, 2014) assumes that the extent to which the user may control the narrative and the story outcome influences identification. Similarly, the Moderated Mediation Model of Media Interactivity (Lin, 2013b) proposes that interactivity, i.e., the possibility to control the video game character, positively affects identification with this character. Some empirical data supports this, showing higher identification scores in users who played rather than watched a video game (Hefner et al., 2007), or in participants who were able to select the plot (rather than merely consuming it) in an audiovisual fiction (Soto-Sanfiel et al., 2010). Other research (Lin, 2013a), however, did not find said associations. Generally, it is important to note that a consistent definition of interactivity is complicated by the fact that interactivity in a written narrative is different from interactivity in a video game (Green & Jenkins, 2014). While in text-based narratives, interactivity pertains to the way in which the reader engages with the narrative (Green & Jenkins, 2014), in video games it involves customizing the character's appearance and willfully directing its actions (Bachen et al., 2016;Birk et al., 2016;Christy & Fox, 2016).
It has been argued that interactive fiction (IF) may be more suitable than video games for answering the question of whether and how interactivity impacts identification. IF is comprised of text-based stories which allow the readers to (ostensibly) influence the direction of the plot at key decision points (Green & Jenkins, 2014). As such, IF is assumed to be similar to a video game, yet entail fewer potential confounds than those found in video games, such as being able to customize your avatar, or experiencing embodiment in immersive virtual environments (see Murphy, 2004). Here, using IF instead of video games in research allows to test for the hypothesizes associations whilst controlling for confounds (see Sherrick et al., 2014;Steinemann et al., 2017). Given their rigor, experiments with IF are better suited than video games to validate the relations between different variables (e.g., interactivity, transportation, immersion) and identification as postulated by theoretical models such as the one of Green and Jenkins (2014). To date, research which relates these constructs under controlled conditions is still rather scarce.
Apart from interactivity, narrative transportation may be regarded another predictor of identification. Broadly speaking, transportation encompasses the cognitive and emotional immersion in a story (Green & Brock, 2000;Green & Jenkins, 2014). A concise conceptualization is an open issue, and different terms are currently used to describe it, such as immersion (Moore & Green, 2020), engagement (Busselle & Bilandzic, 2008), absorption (Slater & Rouner, 2002), and involvement (Klimmt & Vorderer, 2003). Generally, transportation refers to the narrative as a whole, while the conceptually different phenomenon of identification pertains solely to the character(s) within the narrative (Green & Jenkins, 2014). Several researchers (e.g., Behm-Morawitz & Mastro, 2009;Christy & Fox, 2016;Green & Brock, 2000) have hypothesized that transportation into the narrative is a prerequisite for character identification (see also Slater & Rouner's, 2002, theoretical model), yet empirical sustenance of this assumption is still scarce, and more studies are needed to shed light on the hypothesized associations.
Another predictor of identification is thought to lie in the construct of immersive tendencies. Contrary to transportation, which describes the factual immersion in the story, immersive tendencies encompass an individual's predisposition to become cognitively and emotionally involved in a story in the first place (Green & Jenkins, 2014;Mazzocco et al., 2010;Witmer & Singer, 1998). As such, it is an individual trait, which has often been used interchangeably with the construct of transportability, which also describes a person's ability to be engaged with a narrative (e.g., Moore & Green, 2020;Skarbez et al., 2017). While past results support a relationship between transportability/immersive tendencies and identification for narrative video games (Christy & Fox, 2016), the effect of immersive tendencies in the context of interactive narratives like IF remains to be studied (Green & Jenkins, 2014).
Finally, users' gender and age may also be of importance in identifying with media characters. Cohen (2001) hypothesized that diverse demographic variables may influence the identification process by eliciting feelings of similarity between the user and the media character. Accordingly, Slater and Rouner (2002) have included homophily as a predictor in their model. However, past research has not yet sufficiently considered these variables. Only few contradictory results exist, some showing no connection (Cohen et al., 2018;Hall, 2020), while others imply that females identify more strongly with their own gender, and males identify equally with either gender (Hook, 2019).

Study Objective
On the one hand, past research is characterized by heterogeneous definitions and operationalizations of key constructs (e.g., Cohen, 2001), and on the other hand, it suffers from a lack of comparability of findings due to the use of different media (i.e., video games vs. text-based narratives) with varying confounds (e.g., influence of customization on identification). Also, the downfall to using video games when studying identification are the manifold factors which may potentially influence character identification and hence, impede validation of according theoretical models of identification. Hence, the current study set out to examine the postulated associations in a text-based fiction with purported interactivity. It was hypothesized that user's immersive tendencies, their experience of actual transportation into the narrative, and the interactivity of the narrative (active vs. passive) would predict identification with the narrative's main character. Moreover, we included gender and age as possible predictors of character identification to test for their influence.

Stimulus Material
An IF was created using Twine (http:// twine ry. org) which is an open-source software allowing the programming of interactive games. The game used for this study was purely text-based, yet throughout its course, a static picture of the main character was visible on the right-hand side of the text. Two portraits of young Caucasian adults (one male and one female) were created using Adobe Fuse CC and rendered with Octane Render. Both characters were designed to resemble each other with regards to their external features; hence, both were light-skinned, had brunette hair, and wore a brown jacket (see Fig. 1). The IF created for the purpose of this study included a murder case, which was told from the first-person perspective of the detective who was called to the scene to interrogate potential suspects (total word count: 3490 words in German). The game was activated via a link and was adapted to run in all standard browsers (i.e., Internet Explorer/Edge, Mozilla Firefox, Google Chrome, Safari). In the active condition, participants were instructed that they would be playing a game as a detective whose goal was to successfully solve a murder case. Also, participants were told that to achieve this goal, they would have to make several decisions, which would all critically influence the outcome. In contrast, participants in the passive condition were merely asked to read the story about a detective solving a murder case. In both conditions, participants were provided with the detective's last name, Kummer, and with her or his static portrait. Each participant was randomly assigned either to the story with the male main character or to the one with the female main character.
Overall, the course of the narrative was comparable in both conditions, with all subjects ending up reading all information and arriving at the same end of the story, where the detective arrests the murderer. The only difference was that participants in the passive condition clicked on a continue button, while participants in the active condition could choose from two to three options (see Fig. 2). As such, the active condition included real, active choices, whereas the passive condition did not. Hence, there was a factual difference between the two conditions regarding the interactivity of the medium. This is in line with the assumption that while classical text-based narratives are passively consumed, interactive narratives entail active engagement (Green & Jenkins, 2014). Presenting the same plot and ending across conditions, allowed us to reduce possible interfering influences of condition-dependent storylines and endings. This is a method commonly used in studies in the field of interactive fiction (see Jenkins, 2014;Soto-Sanfiel et al, 2010;Steinemann et al., 2015). Overall, participants had to click "continue" or choose from different options 24 times; choices in the interactive version were not recorded.
To assess whether there were any differences between the male and female images in terms of attractiveness, sympathy, and estimated age, and to evaluate whether the interactive narrative was engaging, and the passive and active condition were perceived as such, a pre-study was conducted with 10 young adults aged between 25-30 years who were randomly assigned to either the male or female character and to either the active or passive condition. There were no significant differences in how the male and female avatars were evaluated: Both were estimated to be about 25-30 years old; both were reported to be equally attractive and sympathetic. Also, the main character's gender was always correctly identified by all pre-study participants. With regards to the two narratives, the active condition was judged as more interactive and participants stated they had the impression they could influence the outcome of the story and the main character's decisions.

Procedure
The study was administered in two stages to avoid bias due to asking about behaviors and assessing them in the same session (see Sherrick et al., 2014, for a similar procedure). The first part of the study contained the pre-assessments (demographic screening), and the second part contained the experiment and the post-measures (all remaining questionnaires, see Measures section). All self-report questionnaires were administered online via SoSciSurvey (https:// www. sosci survey. de). The experiment was carried out in strict accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki). Participants were informed about the procedure, duration, and purpose of the study, as well as about voluntary participation and anonymization, and had to consent to these terms before proceeding. Subsequently, participants were asked to provide their e-mail address by means of an opt-in question, which they had to confirm in an e-mail sent to them. To ensure data protection and anonymity, e-mail addresses were stored in a separate dataset on SoSciSurvey and could not be matched to the answers in the questionnaire. After confirming their e-mail, participants were automatically redirected to the first part of the study and answered the pre-questionnaires which entailed the demographic screening (see Measures section) and took about 2-5 min. Afterwards, subjects were asked to generate a personalized 6-digit-code to allow for matching their answers during the second stage. Four days after completion of the pre-assessment, a link was sent to participants via e-mail leading them to the IF (the link was valid for two days to ensure a similar timeframe for the post-responses). Participants were randomly assigned to either the active or passive condition; both took about 20-25 min to be completed and were followed by post-questionnaires (see the remaining questionnaires in the measures section). Finally, subjects were asked to enter their personalized code and were able to choose whether they wanted to receive further information about the results.

Measures
Demographic Data A short demographic survey assessed participant's age, gender, educational level, as well as their reading and computer gaming habits, using a single item each: "How many books do you read in a month?", "How many hours per week do you play computer games?".
Fuchs as their answer were included in the analysis.

Immersive Tendencies
To assess immersive tendencies as a valid indicator of participants' trait transportability, the Immersive Tendencies Questionnaire (ITQ; Witmer & Singer, 1998) was used. For the current study, it was translated into German using the back-translation method (see Brislin, 1970). Also, the first 14 out of the 18 original items were included which assess the individual predisposition to become immersed in a medium on an 8-pt-Likert scale (1 = never to 8 = always). The mean score in our sample amounted to 63.79 (SD = 15.27). The internal consistency (Cronbach's α) for the total scale was good with α = 0.81.
Transportation A German version of the Transport Narrative Questionnaire (TNQ, Green & Brock, 2000) assessed the extent to which participants were transported into and absorbed by the narrative using 15 items (e.g., "While I was reading the narrative, I could easily picture the events in it taking place") on a 7-pt-Likert scale (1 = not at all to 7 = very much). Again, we used the back-translation method to develop the German version of the questionnaire. As proposed by the authors (Green & Brock, 2000), we used the full scale (sum score) for further analyses, which provided a good internal consistency (Cronbach's α = 0.84). The full scale also included 4 items (12-15) pertaining to the impression the participants had of 4 characters appearing in the murder case ("As I read the story, I had a vivid impression of…": detective Kummer (main character), Mrs. Schwarz (wife of the murder victim), Mr. Fuchs (murderer), Mrs. Graf (neighbor). Our sample had a mean score of 59.69 (SD = 13.66) for the total scale.

Identification
The back translated German version of the 14-item Identification with Characters (EDI; Igartua, 2010;Igartua & Páez, 1998) was used to measure the extent to which a person loses self-awareness and thus, has the impression of becoming the character (e.g., "I identified with the character."), and feels cognitive and emotional empathy towards the character (e.g., "In understood the character's feelings or emotions.") on a 5-pt-Likert scale (1 = not at all to 5 = very much). The mean score in our sample was 39.54 (SD = 11.77), the internal consistency for the total scale of 14 items was α = 0.92.

Data Analyses
Data was analyzed using IBM SPSS version 26 (SPSS, Inc. Chicago, USA), considering a significance level of p < 0.05. Power calculations were performed using G*Power (Version 3.1.). Aggregated data are stored on a repository of the Open Science Framework (https:// doi. org/ 10. 17605/ OSF. IO/ VKUPB).

Sample
Participants were recruited via designated German Facebook pages and groups, as well as in gaming related German speaking online-forums. Inclusion criteria were at least 15 years of age. Exclusion criteria, in turn, were participation in the pre-evaluation of the stimulus material.
The pre-assessment was completed by N = 424 individuals, of which n = 304 also participated in the second stage. Among the 304 data sets, n = 67 had to be excluded because the personal codes could not be matched (and thus the pretest and post-test could not be paired). After excluding data sets with missing responses (< 20% missing) or conspicuous response patterns (e.g., exclusively ticking the middle category) as well as with outliers (standardized residuals beyond ± 2.5), a total of N = 228 participants were included in the current analyses (see Fig. 3, CONSORT Flow Diagram). A posteriori power calculation was performed. With a sample of N = 228 and an achieved significant result, a power (1 -β) of 0.99 was realized. Hence, it can be stated with over 99% confidence that effects of size f 2 = 0.22 or greater exist.
Participants were randomly allocated to either the active condition, n = 122 (54%), which could choose from different options, or the passive condition, n = 106 (47%), which could only click on the continue button. The adherence check showed that all n = 228 participants correctly answered the question of who had murdered Mr. Schwarz and hence, all were included in further analyses.
Participants' age ranged from 15 years to above 65 years, with the largest proportion being 20 to 29 years old (54%). Overall, n = 168 (74%) of the participants were female and n = 60 (26%) identified as male. Almost half of the sample (47%) reported a university degree as their highest educational qualification. In addition, n = 160 (70%) of participants reported to engage in two or less hours in video gaming per day. Furthermore, n = 129 (57%) participants read one to two books per month, n = 52 (23%) read three to five books, and n = 28 (12%) read more. A small number Assessed for eligibility (n=424) Excluded (n=120) only answered pre-assessment (n=120) Analysed ( To exclude the possibility that in one of the two conditions, despite randomization, there was a disproportionately large accumulation of the following (possibly influencing) characteristics, group differences between the active and passive condition were analyzed using univariate ANOVA: Age (p = 0.998), educational qualification (p = 0.760), computer habits (p = 0.875), and reading habits (p = 0.714) did not yield differences between the active and the passive condition. Concerning gender, there were marginally more male participants in the active condition (29%) than in the passive one (23%), which resulted in non-significant group differences (p = 0.341) using the Wilcoxon rank-sum test. The Person-Chi Square test supports this assumption (χ 2 (1, N = 230) = 0.912, p = 0.340).

Multiple Linear Regression Analysis
Prior to the multiple regression analysis, normality tests were performed and confirmed a normal distribution of the included metric variables. Linear interrelationships between the predictors and the outcome as well as homoscedasticity were graphically verified and the data met the assumption of independent errors (Durbin-Watson value = 1.68). The centered leverage criterion [0.006, 0.096] and the Cook-distance criterion [0.000, 0.081] showed no outliers after excluding two outliners prior to the calculation of the regression. Intercorrelations of the three included predictors ranged between -0.13 and 0.34. The data indicated that multicollinearity was not a concern (interactivity: tolerance = 0.96, variance inflation factor [VIF] = 1.03; narrative transportation: tolerance = 0.85, VIF = 1.17; immersive tendencies: tolerance = 0.81, VIF = 1.24).
A multiple linear regression analysis (method: forced entry) was used to test whether interactivity (active vs. passive condition), narrative transportation (TNQ), immersive tendencies (ITQ) significantly predicted participants' identification with the story's main character (EDI), as well as age and gender. The multiple regression analysis indicated that 67% of the outcome variance was explained by the model, R 2 = 0.67, F(5, 222) = 91.452, p < 0.001. Interactivity significantly predicted identification (β = -0.26, p < 0.001), as did transportation (β = 0.70, p < 0.001), immersive tendencies (β = 0.09, p < 0.034), and gender (β = 0.08, p = 0.048; though only barely significant, male gender was by trend associated with higher levels of identification). Age group (β = -0.03, p = 0.412) did not show significant effects on identification. When character gender and interaction of character gender and participant gender were included in the regression model, the model remained significant and gender lost its weak but significant impact. Neither predictor (character gender: β = -0.09, p = 0.490; interaction: β = 0.14, p = 0.440) showed a significant effect on outcome, nor did they influence the results of the other predictors (Table 1).

Discussion
Identification with fictional characters -be it in text-based narratives, movies, or video games -is thought to positively shape media persuasiveness, enjoyment, and learning. Hence, understanding what lies at the core of character identification is of interest particularly to those applications that rely on successfully inducing identification for the promotion of social change or health behaviors. Yet, previous research largely suffers from limited cross-study comparability due to differing definitions of key constructs and due to the use of media with varying degrees of interactivity (i.e., video games, film, text-based narratives). Hence, the current study set out to add to the body of data by evaluating the influence of media interactivity and user variables Total N = 227; ITQ = Immersive Tendencies Questionnaire; TNQ = Transport Narrative Questionnaire; CI = confidence interval; LL = lower limit; UL = upper limit a 1 = active condition, 2 = passive condition. b 1 = < 15 years; 2 = 15 to 19 years; 3 = 20 to 24 years; 4 = 25 to 29 years; 5 = 30 to 34 years; 6 = 35 to 39 years; 7 = 40 to 44 years; 8 = 45 to 49 years; 9 = 50 to 54 years; 10 = 55 to 59 years; 11 = 60 to 64 years; 12 = 65 years or older. c 1 = female and 2 = male such as trait immersive tendencies, gender, age, and factual transportation into the narrative on character identification in a text-based fiction. Results show that apart from age, all predictors included in the multiple linear regression model were significant. In other words, identification with the story's main character, detective Kummer, was predicted by participants' gender, their trait immersive tendencies and experienced transportation into the narrative, as well as the story's interactivity.

Interactivity and Identification
The possibility to select from several choices within our text-based narrative positively and significantly impacted identification with the story's main character. This finding is in line with past studies (Hefner et al., 2007;Soto-Sanfiel et al., 2010), and it supports theoretical models such as the Conceptual Model of Interactivity Effects (Green & Jenkins, 2014) or the Moderated Mediation Model of Media Interactivity (Lin, 2013b). As such, the current research adds empirical sustenance for the key role of media interactivity in the formation of identification. However, one issue remains: while our media was factually interactive in that it afforded choices between different options, the storyline and outcome were the same across conditions (like in Jenkins, 2014;Soto-Sanfiel et al., 2010;Steinemann et al., 2015). The current results support the success of the manipulation, yet, it is possible that some participants may not have believed that their choices changed the story's outcome. Future research should carefully control for whether and which participants may find out about the manipulation and react with frustration and/ or disappointment.
Another issue is the question of whether the findings are also valid for other media types. While IF is considered well suited for studying the link between interactivity and identification (see Sherrick et al., 2014;Steinemann et al., 2017), findings based on IF-research designs may not be completely comparable to video games. The visual richness and action dynamics of video game usually exceed those of IF by far. Video games entail not only choosing, designing, and customizing one's avatar with regards to her or his appearance and skills, but they also rely on perceiving the three-dimensional environment from a fist person perspective and on performing directed, willfully steered actions (Bachen et al., 2016;Birk et al., 2016;Christy & Fox, 2016).
Here, the concept of agency seems to be of relevance: It describes a sense of being the cause of the media character's actions (Fox et al., 2015). As such, agency has also been associated with the experience of embodiment (see Murphy, 2004), which, in a basic sense of the concept, may be described as having control over the body of the video game character (Aymerich-Franch, 2020). Hence, in contrast to IF, where interactivity mainly relies on deciding about the direction of the plot, in video games, interactivity may entail a more multi-faceted concept. As stated by Birk et al. (2016), "while traditional narratives rely on empathy […] -the viewer's identification with the emotions of a character -interactive computer games emphasize agency -the direct control of a player over the behavior and development of their character" (p. 2983). In this sense, media with differing levels of interactivity and action options may tap into substantially different aspects of identification: While IF may prompt empathizing with a character (feeling with the character), video games may rather lead to a shift of selfperception and to the impression of becoming the character (Klimmt et al., 2009(Klimmt et al., , 2010.

Transportation, Immersive Tendencies, and Identification
In our study, the experience of being transported into the narrative significantly predicted identification with the story's main character, adding sustenance to past findings (see Behm-Morawitz & Mastro, 2009;Christy & Fox, 2016;Green & Brock, 2000;Slater & Rouner, 2002). An open issue, however, is the type of narrative used. Generally, it is assumed that the quality of the text and the story's genre may impact transportation (Green & Brock, 2000). A well-written story is thought to evoke higher levels of immersion into the narrative than a poorly crafted story (Mazzocco et al., 2010). Accordingly, an exciting storyline and emotionally engaging content -like in adventure novels or horror stories -is expected to be more transporting than a mundane story with little excitement or emotional relevance. To assure comparability with prior studies (see Sherrick et al., 2014, for a similar approach), we developed a murder story within which participants assumed the role of a detective. While a detective narrative is expected to be engaging, it cannot be ruled out that the quality of writing (i.e., language level) may lead to a different cognitive and emotional engagement.
Apart from narrative content, the type of medium used may also be of essence for the experience of transportation. In contrast to, for instance, video games, texts are thought to rely on more imaginative investment from the recipient (Green et al., 2004). By implication, those individuals who have a higher likelihood of becoming cognitively and emotionally immersed in media are also expected to experience more transportation when reading a story (Christy & Fox, 2016;Green & Jenkins, 2014). As such, an individual's immersive tendency is regarded highly predictive of narrative transportation (Dal Cin et al., 2007;Mazzocco et al., 2010) and, thus, also for identifying with the narrative's character (Christy & Fox, 2016;Dal Cin et al., 2007). The current experiment was able to confirm this by showing that individuals with higher immersive tendencies also reported more identification with the detective. However, it must be noted that, among the predictors included in the multiple linear regression model, the factual immersion, i.e., the transportation into the narrative (as assessed by the TNQ), was a stronger predictor (β = 0.70) of identification than immersive tendencies, i.e., the predisposition to become immersed in a medium (as assessed by the ITQ; β = 0.09); transportation into the narrative was the strongest of all included predictors of the model. In future studies, the effect of immersive tendencies or transportation should be examined alone, alongside controlling for the confounding influence of the narrative's interactivity, content and quality of writing.

Gender, Age, and Identification
In the face of limited data, demographic variables were also included in the regression model, however, age did not emerge as a significant predictor of identification. Given the broad age range (15 to 65 years) of our sample, the non-significant finding implies that identification may occur independent of users' age. A confounding factor, however, may be that our largest age group was between 20-29 years (approx. 50% of the sample) and as such, matched the detective's age of 25-30 years as estimated in the pre-study. But to conclude that demographic similarity had no impact on identification (see Cohen, 2001;Slater & Rouner, 2002) is not feasible considering our study design.
Contrary to age, gender predicted identification with the story's main character. Given that this predictor was barely significant, it is difficult to make reliable statements about which gender had which effect. We may at this point only cautiously state that in our sample -and in contrast to past findings (Hook, 2019) -males tended to report higher levels of identification. Furthermore, when we included character gender and interaction of character and participant gender (as an indicator of demographic similarity between participants and the characters) in the regression model, the model remained significant but gender lost its significant impact (which was weak to begin with). Therefore, we were not able to confirm Hook's (2019) finding that females identify more strongly with their own gender, while males identify with both genders.
In sum, more research is needed to add further sustenance to the question of whether and how gender influences identification. Studies should strive for a 50:50 gender ratio and for balanced age groups, and include age as a continuous variable (in the current study, age was collected using categories), as well as control for perceived demographic similarity with media characters.

Limitations and Conclusion
The strength of the current research lies in the use of widely applied, comprehensive measures for the assessment of key variables, as well as in closely adhering to theoretical models of media interactivity and identification (see Green & Jenkins, 2014;Lin, 2013b;Slater & Rouner, 2002). Furthermore, we followed the recommendation to use IF as a means of studying said associations, because this reduces the risk of confounds (Sherrick et al., 2014) and we included gender and age as possible predictors (Cohen, 2001). As such, we were able to fill a gap in the body of data. However, there are also several limitations to the current study.
In accordance with Sherrick et al. (2014), the story's main character was assigned to participants. Hence, they could not choose between the male or female detective, let alone customize him or her. Customization, however, is assumed to potentially impact whether and to what extent users identify with the fictional character (see Birk et al., 2016) and should be regarded in future research. Furthermore, we used a cross-sectional design. Yet, there is evidence that time may be of essence: For instance, Lin (2013a) reported that although their participants empathized with a video game character, the duration of game play did not seem to be long enough to evoke more than emotional empathy. In support of this, a study (Van Reijmersdal et al., 2013) found that identification intensified with longer gaming time. Longitudinal designs are needed to shed more light on this issue. Furthermore, we used static pictures to depict the detective. Other IF-studies resorted to animated pictures (see Sherrick et al., 2014) or did not use any illustrative material at all (see Green & Jenkins, 2014). Although we preevaluated the pictures to ensure that they were comparable across genders and did not elicit any negative responses, completely dropping them could have had a different effect on identification. In general, many interactive narratives are devoid of the rich character description found in traditional narratives (books etc.); as such they leave more space for the readers' imagination (Green & Jenkins, 2014). Yet, whether this lack of concrete character information facilitates putting the recipient into the shoes of the character, remains to be answered in future studies. Also, the type of recruitment we chose for the current study and the resulting sample may be regarded a limitation. Given that IF may be used to foster education and health (Steinemann et al., 2017), appropriate target groups (i.e., children and adolescents or persons at risk of developing a mental disorder) should be taken into account in future studies.
Finally, in order to increase experimental control, we used a purportedly interactive story (see Jenkins, 2014;Soto-Sanfiel et al., 2010;Steinemann et al., 2015). Our active condition included real, active choices, but in fact the story content and ending were the same as in the passive reading-only condition. Hence, while the medium was factually interactive (in that it allowed making choices), the possibility remains that users may not have perceived it as such. Although a small pre-study confirmed the success of this manipulation, more in depth assessment of participants' subjective experiences would allow for a more differentiated assessment of interactivity (and possible the feeling of being deceived). Hence, future studies should include direct evaluations of user experiences.
Hence, the main challenge of upcoming research in the context of IF will be to simultaneously regard a range of key factors in the study of character identification, like customization vs. assignment of characters, similarity vs. differences, and presence vs. absence of illustrative material (i.e., pictures of the character) and to evaluate these aspects in different samples.
Funding Open access funding provided by Medical University of Vienna. No funds, grants or other support was received for conducting this research.
Data availability statement Data is available upon request from the corresponding author.

Conflict of interest statement
The authors declare that they have no conflict of interest.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.