The role of prior attitudes in narrative persuasion: Evidence from a cross-national study in Germany and the United States

ABSTRACT

Narrative messages are assumed to be more effective in changing recipients' attitudes than non-narrative messages. However, empirical evidence to support this assumption is sparse. We incorporated theoretical assumptions about the mechanisms of narrative persuasion into the two-step model of defensive processing to test whether narratives were more effective in changing recipients' attitudes toward legalizing marijuana for recreational use. We conducted two parallel experiments, one in Germany (N = 157) and one in the United States (N = 399). Our findings did not support the general assumption that narratives were more effective than non-narrative messages. However, prior attitudes were identified as an important unique factor in shaping recipients' transportation and identification and, in turn, the recipients' attitudes.

KEYWORDS:

• Narrative persuasion
• prior attitudes
• narrative messages
• transportation
• identification
• replication

Over the past two decades, narrative messages have attracted the interest of communication scholars in various fields, including health communication, media psychology, political communication, and journalism studies (Braddock & Dillard, 2016; Früh & Frey, 2014; Green et al., 2002). We employed Kreuter et al.’s (2007) definition of a narrative as “a representation of connected events and characters that has an identifiable structure, is bounded in space and time, and contains implicit or explicit messages about the topic being addressed” (p. 222). The notable finding that narrative messages can affect recipients’ beliefs, attitudes, intentions, and behaviors has sparked considerable scholarly interest (Braddock & Dillard, 2016). Their persuasive impact has even been found to exceed the effects of non-narrative messages (Oschatz & Marker, 2020; Shen et al., 2015), messages that comprise “expository and didactic styles of communication that present propositions in the form of reasons and evidence supporting a claim” (Kreuter et al., 2007, p. 222). For example, studies have shown that recipients of narrative messages reported stronger message-consistent health promoting attitudes towards preventive care (Martínez Martínez et al., 2018) and intentions to quit smoking (Kim et al., 2012) than recipients of non-narrative messages. Narratives have also outperformed non-narrative messages in evoking positive attitudes toward vulnerable groups such as people with disabilities (Wong et al., 2017), immigrants, and the elderly (Oliver et al., 2012). These effects have been observed for long-form narratives and also short stories or even 30-s public service announcements that embed narratives within strategic messages (Dahlstrom et al., 2017; Kim et al., 2012; Oliver et al., 2012). Whereas multiple empirical studies examined the persuasive effects of narrative in contrast to non-narrative messages (see, for example, Shen et al., 2015), we know comparably little about the boundary conditions of such persuasive effects. This study shed light on this gap in research.

In general, a persuasive impact can be differentiated into shaping, reinforcing, or changing recipients’ beliefs, attitudes, intentions, and behaviors (Perloff, 2017). Shaping refers to the formation of, for example, attitudes toward objects, individuals, or groups. Reinforcing implies the affirmation of existing attitudes. Changing refers to the modification of recipients’ prior attitudes. Theorists have argued that narrative messages are particularly effective compared to non-narrative messages in changing recipients’ attitudes “if the persuasive content is counter-attitudinal” (Slater & Rouner, 2002, p. 173), that is, in reception situations where recipients come across messages with arguments, information, and/or positions that are inconsistent with their prior attitudes (Dal Cin et al., 2004; Slater & Rouner, 2002). However, whether narrative messages are more powerful than non-narrative messages in changing recipients’ prior attitudes has hardly been examined in narrative persuasion research, as only a few empirical studies have assessed recipients’ baseline attitudes (e.g., Wojcieszak & Kim, 2016; Zhao & Fink, 2020). This study addressed this question in depth and thus contributed to an understanding of the boundary conditions of the persuasive impact of narrative and non-narrative messages. We answered the research question in a cross-national experimental study, conducted in Germany and the USA, respectively, by manipulating message type (narrative vs. non-narrative) and consistency of the message with the recipients’ prior attitudes (consistent vs. inconsistent). Although replication is considered an important part of social scientific research (Dienlin et al., 2021), it is rarely conducted in communication science (McEwan et al., 2018). Hence, we chose a cross-cultural replicative design as Tiokhin et al. (2019) point to the importance of replicating findings across diverse cultural populations to improve the generalizability of the results.

Mechanisms of narrative persuasion

The persuasive advantage of narrative over non-narrative messages has been ascribed to their unique ability to capture their audiences. This state is termed narrative involvement – “the interest with which viewers follow the events as they unfold in the story” (Moyer-Gusé, 2008, p. 409). Narrative involvement can be differentiated into involvement with the story and involvement with characters (Moyer-Gusé, 2008). Involvement with the story has also been referred to as transportation (Green & Brock, 2000), narrative engagement (Busselle & Bilandzic, 2009), and absorption (Slater & Rouner, 2002). These concepts share the common notion that recipients focus more on how the story unfolds than on their surrounding environment (Moyer-Gusé, 2008). Transportation, “a convergent process, where all mental systems and capacities become focused on events occurring in the narrative” (Green & Brock, 2000, p. 701), has been used most frequently to conceptualize and measure narrative involvement with the story (Oschatz & Marker, 2020; Tukachinsky & Tokunaga, 2013). In an intense state of transportation, recipients are thought to develop cognitive and affective connections with the story and conjure vivid images of the story’s events (Appel et al., 2015; Cohen, 2001; Green & Brock, 2000).

Involvement with characters is an overarching category of concepts that describes how viewers interact with characters, such as identification (Cohen, 2001) and parasocial interaction (Horten & Wohl, 1956). In our study, we relied on identification, a “process that culminates in a cognitive and emotional state in which the audience member is aware not of him- or herself as an audience member, but rather imagines being one of the characters in the text” (Cohen, 2001, p. 252), as one of the most prominently used concepts in narrative persuasion (de Graaf et al., 2016). In an intense state of identification, recipients process knowledge from the perspective of the character and share motivations, emotions, and goals (Cohen, 2001).

In such an intensive state of transportation and identification, narrative messages are perceived as less patronizing, threatening, and/or not even having persuasive intent in contrast to non-narrative messages. Consequently, the recipients develop attitudes and intentions that converge with the position provided in the message (Knowles & Linn, 2004b; Moyer-Gusé, 2008; Slater, 2002). Indeed, studies have found that recipients of narrative messages tend to experience greater transportation and identification with the message than do recipients of non-narrative messages (Kim et al., 2012; Oschatz et al., 2019). Greater transportation and identification, in turn, was associated with greater persuasive impact in line with the position presented in the story (Kim et al., 2012; Oliver et al., 2012; Oschatz et al., 2019; Shen et al., 2017). In line with the theoretical assumptions and replicating previous studies, we posed the following hypotheses on the direct and indirect effects of narrative in contrast to non-narrative messages on persuasive outcomes.

H₁: Narrative messages will produce stronger message-consistent persuasive effects than non-narrative messages.

H₂: An effect of message type on message-consistent persuasive effects will be mediated by transportation and identification, such that narrative messages will facilitate greater transportation and identification than non-narrative messages (H_2a), and greater transportation and identification will be associated with greater message-consistent persuasive effects (H_2b).

The role of prior attitudes in narrative persuasion

Communication scholars have theorized that narrative messages may be particularly effective at changing attitudes among audiences who hold positions on an issue that are counter to those being advocated in the narrative message (Dal Cin et al., 2004; Slater & Rouner, 2002). Other scholars have theorized about how people process counter-attitudinal messages, regardless of whether they appear in narrative or non-narrative form. Here, we attempted to synthesize these perspectives to make specific predictions about the role of transportation and identification in processing counter-attitudinal content.

Giner-Sorolla and Chaiken (1997) developed a two-step model of defensive processing based on the heuristic-systematic model (Chaiken, 1980; Chaiken et al., 1989) and motivated reasoning (Kunda, 1990). According to the model, recipients differ in the processing of messages depending on whether the message is consistent or inconsistent with the recipients’ prior attitudes. A message consistent with prior attitudes is processed with little mental effort as the recipients are satisfied with the arguments confirming their point of view. When a recipient comes across a message that conveys arguments inconsistent with their prior attitudes, they switch to a defensive processing mode. Defense motivation is the desire to hold on to consistent attitudes (Chaiken et al., 1996). According to the model, inconsistent messages are processed systematically as the recipients invest more cognitive effort in carefully considering and evaluating the arguments of the persuasive message.

Integrating the theoretical assumptions about mechanisms of narrative persuasion into the defensive processing model suggests that both narrative and non-narrative messages that explicitly convey arguments consistent with a recipient’s point of view are likely to be processed heuristically, without critical assessment of its content, and with similar levels of transportation and identification. This in turn should result in forming stronger attitudes (by integrating new favorable arguments) and/or affirming existing attitudes for both narrative and non-narrative messages because, in both cases, the arguments are likely to be accepted without much scrutiny. In this pro-attitudinal scenario, we would not expect large differences in attitudinal change between narrative and non-narrative messages.

Theorizing on transportation and identification suggests that a different process may occur when respondents encounter counter-attitudinal narrative vs. non-narrative messages. In this scenario, we contend that narrative messages should be more likely than non-narrative messages to bypass defensive processing among those with inconsistent prior attitudes by triggering greater transportation and identification for a narrative vs. non-narrative message. Whereas counter-attitudinal content may trigger defense motivation in a non-narrative message, the recipients of a narrative message who enter intense states of transportation are likely to have fewer cognitive resources available to systematically and critically process arguments that are inconsistent with their point of view (Dal Cin et al., 2004; Knowles & Linn, 2004a; Kreuter et al., 2007). Furthermore, as high levels of identification with the characters in a narrative create a sense of connectedness and trust (Slater & Rouner, 2002), the recipients might not want to compromise their pleasurable experience with a critical attitude (e.g., Kreuter et al., 2007) or, at a minimum, are more open to receiving arguments from an opposing point of view. This should in turn increase the likelihood of a narrative modifying the recipients’ prior attitudes toward the issue, resulting in a greater persuasive effect for (initially) counter-attitudinal respondents compared to audiences for whom the content simply affirms previously held attitudes. Following this logic, we derived the following hypotheses for the current investigation:

H₃: The effect of message type on transportation and identification is moderated by the recipients’ prior attitudes, such that the effect of narrative (versus non-narrative) messages on transportation and identification will be stronger among respondents exposed to messages that convey arguments inconsistent with recipients’ prior attitudes.

Method

Research design

We conducted two replicative online field experiments with participants in Germany (N = 157) and the USA (N = 399) in December 2019 and January 2020 using a 2 (message type: narrative vs. non-narrative) × 2 (consistency of the message with prior attitudes: consistent vs. inconsistent) between-subjects design. The goal of this study was to investigate the moderating role of message consistency with prior attitudes on transportation and identification and, consequently, on the persuasive effects of narrative and non-narrative messages. We selected the issue of recreational marijuana legalization. Whereas medical marijuana can be legally prescribed by a physician in both Germany and in more than half of the states within the USA, recreational marijuana is illegal both in Germany and in a majority of the states in the USA. We expected this issue to be particularly useful for our research context for three reasons. First, the issue of marijuana legalization is discussed in both countries with similar arguments in the public debate. Second, public opinion data suggested that there are sizable groups of both opponents and proponents about legalization in both countries, with about half of the population having a tendency to support or oppose the legalization of recreational marijuana in Germany (Infratest Dimap, 2019) and at least a third remaining opposed to legalization in the USA (Pew Research Center, 2019). Third, these same sources of public opinion data showed that perceptions of recreational marijuana legalization have shifted considerably in both countries over time, suggesting that these attitudes are movable. The study was approved by the Institutional Review Board for Human Participants at Cornell University (IRB Protocol ID 1912009253).

Recruitment and participants

There was no reliable effect size available for the interaction of message type and consistency of the message with prior attitudes to determine the required effect size in a power analysis. Consequently, we did not conduct a power analysis prior to conducting the study but instead sought to estimate such an effect size in two independent, cross-national samples. We recruited the German survey respondents on a German university campus for course credit points using convenience sampling. A total of 160 respondents completed the questionnaire. We removed three participants from the analysis due to excessive amounts of missing data and/or rapid questionnaire completion based on an analysis of several quality criteria indicators embedded in the SoSciSurvey survey tool used for data collection (Leiner, 2019). The final sample consisted of 157 students aged 18–47 years (M = 22.1, SD = 3.7). The majority of the participants were females (78%).

We recruited the US survey respondents using Prolific Academic, a professional survey research platform online that recruits US adults to participate in surveys for monetary compensation. A total of 414 respondents completed the questionnaire. We removed 15 participants from the analysis who left several items unanswered and/or rushed through the survey (stimulus message reading time <30 s or survey completion time <3 min). The final sample included 399 respondents aged 22–78 years (M = 43.1, SD = 14.3). Each respondent received 1.25 US dollars based on the average time spent on the survey. Roughly half (52%) were female, the vast majority (86%) identified as white, and most (56%) had a college degree or higher level of formal education.

Stimulus materials

We created the message stimuli and presented the messages as online news articles on the topic of recreational marijuana legalization by the fictitious author Jennifer Schwarz. We first prepared the four news articles in English and then translated them into German. We wrote two articles as narrative and two articles as non-narrative messages. One narrative and one non-narrative news article presented pro-legalization opinions, whereas the other two presented anti-legalization views. We crafted these news articles iteratively with multiple sources of feedback from colleagues and professional journalists to create credible stimuli.

The narrative and non-narrative article versions were identical in title, lead-in, and the first paragraph of the article. The manipulation was established in the second paragraph, which introduced Robert Hoffmann, a self-employed architect and father of two, as the founder of the (pro- or anti-marijuana legalization) citizen initiative “(Don’t) Legalize It.” In the non-narrative articles, Hoffmann provided plain arguments for or against the legalization of recreational marijuana use supported by scientific evidence. In the narrative articles, Hoffmann told a personal story about his motivation. He revealed that he smoked pot himself when he was in his 20s but stopped when his best friend Martin died as a consequence of his marijuana consumption (described further below). The main arguments in the narrative and non-narrative conditions remained constant across conditions, but we peppered the narrative with personal experiences, thoughts, and emotions of Robert Hoffmann. For example, the non-narrative version stated, “Hoffmann believes that his work to oppose/to support legalization of recreational marijuana might prevent others from suffering the same fate,” whereas the narrative version included a reference to his children and his friend Martin. “Hoffmann recently found out that some of his oldest son’s friends had begun experimenting with marijuana. While he knows that all of his efforts won’t bring his friend Martin back, Hoffmann believes that his work might prevent others from suffering the same fate.”

The narrative and non-narrative articles slightly differed in length due to the need to hold arguments constant and to add narrative elements of plot, character, and setting. We ensured that the corresponding German and English versions, as well as the pro- and anti-conditions, were comparable in length.

Arguments for the pro- and anti-legalization conditions were adapted from a previous study that gauged perceived argument strength for a variety of reasons to (not) legalize recreational marijuana (Niederdeppe et al., 2019). We chose pro- and anti-arguments rated as strongest by the study participants and that the author team perceived to be relevant in both German and US public discourse on the legalization of recreational marijuana. Title, lead-in, and the first paragraph were held constant in word count and content. The wording clearly displayed a pro or anti-position toward the issue. For example, the pro-legalization articles lead-in stated that “Many people believe that the benefits of legalization would outweigh any potential harms,” whereas the anti-legalization version read “Many people believe that the harms caused by legalization would outweigh any potential benefits.” In the non-narrative conditions, Robert Hoffmann talked about benefits like “Legalizing recreational marijuana will protect marijuana users from consuming unsafe or tainted marijuana products” or harms of legalization like “Legalizing recreational marijuana use will increase use of other illegal drugs. Youth will try marijuana and then move on to other drugs like heroin or cocaine.” The narrative conditions made use of exactly the same arguments but related them to the fate of Robert’s friend Martin. In the pro-legalization article, Robert explained how Martin was poisoned by the toxic chemicals from contaminated marijuana products and argued that this would not have happened if marijuana was legal. By contrast, in the anti-legalization article, Robert described how Martin gradually moved from marijuana to heroin and eventually died from an overdose. We held other article aspects constant across message conditions. All materials used in this study, including full text of each stimulus article, questionnaires, datasets, and syntax, are available via the open science framework (OSF): https://osf.io/9zfp2/

Procedure

We conducted both experiments online. We invited respondents to participate in a study with the goal of understanding how people respond to media coverage on controversial topics, using the issue of recreational marijuana as an example. At the beginning of the questionnaire and prior to stimulus exposure, we asked participants about their interest in the issue of marijuana legalization, their (prior) attitude toward legalization (support/oppose), and how certain they were about their position toward recreational marijuana legalization. We then randomly assigned them to one of the four message conditions. In the German sample, 39 students were in the narrative pro-legalization condition, 39 in the narrative anti-legalization condition, 39 in the non-narrative pro condition, and 40 in the non-narrative anti-legalization condition. In the US sample, 101 students were in the narrative pro condition, 99 in the narrative anti condition, 100 in the non-narrative pro condition, and 99 in the non-narrative anti condition. After reading, participants answered a series of questions that included measures for transportation, identification, and post-exposure attitude toward marijuana legalization.

Measures

Manipulation checks

We asked respondents to rate the stimulus messages using a 7-point semantic differential scale. The respondents indicated the perceived message type (1 = not a personal story, 7 = a personal story) and message direction (1 = against legalization, 7 = for legalization). A two-way Analysis of variance (ANOVA) showed an effective manipulation of the message type (Table 1) and message direction in both samples (Table 2). Respondents indicated that the narrative articles were more of a personal story (Germany: M = 6.27, SD = 1.11; USA: M = 6.22, SD = 1.19) than the non-narrative articles (Germany: M = 4.04, SD = 1.70; USA: M = 3.85, SD = 1.73). Moreover, respondents indicated that the pro-legalization article was more supportive of the legalization of recreational marijuana use (Germany: M = 6.53, SD = 1.27; USA: M = 6.68, SD = 0.91) than the anti-legalization article (Germany: M = 1.47, SD = 1.36; USA: M = 1.41, SD = 1.15).

Table 1. Results of the two-way ANOVAs with message type and message direction as independent variable and rating as personal story as dependent variable.

IV	Sum of squares	df	F	p	η^2	95% CI
Group 1 = Germany
Type	195.4	1	93.45	<.001	.38	[0.26, 0.48]
Direction	0.1	1	0.05	.824	<.001	[0.00, 0.02]
Type * direction	0.2	1	0.11	.740	<.001	[0.00, 0.03]
Group 2 = USA
Type	555.9	1	250.76	<.001	.39	[0.32, 0.45]
Direction	0.8	1	0.378	.539	<.001	[0.00, 0.01]
Type * direction	0.0	1	0.014	.907	<.001	[0.00, 0.01]

Table 2. Results of the two-way ANOVAs with message type and message direction as independent variable and rating of perceived message direction as dependent variable.

IV	Sum of squares	df	F	p	η^2	95% CI
Group 1 = Germany
Type	0.8	1	0.44	.508	<.001	[0.00, 0.03]
Direction	1003.5	1	575.85	<.001	.79	[0.74, 0.83]
Type * direction	0	1	0.02	.884	<.001	[0.00, 0.01]
Group 2 = USA
Type	0	1	0.02	.883	<.001	[0.00, 0.00]
Direction	2767.6	1	2567.68	<.001	.87	[0.85, 0.88]
Type * direction	0	1	0	.986	<.001	[0.00, 0.00]

Prior attitudes and attitude certainty

We measured prior issue attitudes by asking respondents to what extent they opposed or supported legalizing marijuana for recreational use on a 7-point scale (1 = strongly oppose, 7 = strongly support). We used a single-item measure to minimize the extent to which we primed recipients to think about the issue prior to exposure to the stimulus in our pretest-posttest design. This measurement was also in line with recent studies that assessed prior attitudes to marijuana legalization using a single-item measure (e.g., Barker & Moreno, 2021; Ellis et al., 2019; Smith & Merolla, 2020). Attitude certainty was measured on a 7-point scale (1 = very uncertain, 7 = very certain) asking how certain the respondents were regarding their expressed position toward the legalization of recreational marijuana. In contrast to our expectations of split public opinion towards legalization, respondents in both samples were predominantly in favor of legalization (German sample: M = 5.05, SD = 1.85; US sample: M = 5.41, SD = 2.03). Moreover, the vast majority of the respondents indicated a high level of certainty of their (prior) attitude toward the issue (German sample: M = 5.24, SD = 1.62; US sample: M = 6.14, SD = 1.28). We discuss the implications of these observations in the limitations section.

Message consistency with prior attitudes

We used the respondents’ prior issue attitudes (1 = strongly oppose, 7 = strongly support) and the leaning of their experimental condition (pro vs. anti-legalization) to determine to what extent the stimulus message was consistent with respondents’ personal point of view. The respondents’ prior attitudes in the anti-legalization conditions were recoded so that higher values corresponded with greater consistency of prior attitudes with received messages. For example, if a recipient strongly opposed the legalization of recreational marijuana (e.g., scored 1 on the prior attitude scale) and then read an anti-legalization message, the received message was recoded as highly consistent (=7) with the respondent’s personal view. Thus, consistency with prior attitudes was captured on a scale ranging from (1) highly inconsistent to (7) highly consistent with the respondent’s prior attitudes (German sample: M = 3.87, SD = 2.12; US sample: M = 4.08, SD = 2.47).

Persuasive effects

We measured persuasive effects of the messages based on recipients’ post-exposure attitude toward recreational marijuana legalization. Post-exposure issue attitudes were assessed with the same measure as the recipients’ prior attitudes. After reading the article, we asked respondents on a 7-point scale (1 = strongly oppose, 7 = strongly support) the extent to which they now opposed or supported legalizing marijuana for recreational use. Post-exposure issue attitudes were then recoded based on the received direction of the message: if a recipient received an anti-legalization message, the response to the post-issue attitude message was recoded so that higher values of the recipients’ post-test issue attitude represented stronger attitudes in line with the position provided in the message (German data: M = 4.17, SD = 2.09; US data: M = 4.21, SD = 2.49).

Mediators: transportation and identification

We assessed respondents’ transportation with five items from the Transportation Scale–Short Form (Appel et al., 2015) on a 7-point Likert-type scale (1 = not at all, 7 = very much) (e.g., “I was mentally involved in the article while reading it” and “While reading the article, I had a vivid image of Robert Hoffmann”). Identification was measured using 10 items taken from Cohen (2001) on a 7-point Likert-type scale (1 = not at all, 7 = very much). We designed both the narrative and non-narrative messages to feature a character (Robert Hoffman) but added additional details on plot and setting in the narrative message. This decision enabled us to measure identification with Robert Hoffmann in both the narrative and non-narrative messages (e.g., “I was able to understand the events in the article in a manner similar to that in which Robert Hoffmann understood them” and “I think I have a good understanding of Robert Hoffmann”).

Next, we established measurement invariance on a configural, metric, and scalar level (Meitinger et al., 2020; Pirralha, 2020) for the latent constructs of transportation and identification across both samples to ensure comparability of results across countries using multi-group confirmatory factor analysis (MGCFA). We collapsed the two datasets for this purpose. Following Weiber and Mühlhaus (2014), we conducted an exploratory factor analysis using principal axis factoring (PAF) prior to the MGCFA to explore unidimensionality of the measures and to assess whether all manifest items loaded on their respective latent construct. The PAF (KMO = .940, Bartlett test p < .001) suggested a two-factor solution that largely corresponded to the items from the original transportation and identification scales. However, two of the identification items, “While reading the article, I felt as if I was part of the action” and “While reading the article, I forgot myself and was fully absorbed,” loaded on the factor that included all of the transportation items. This was plausible as the two identification items correspond with the absorption dimension described by Cohen (2001) that, similarly to transportation, captures a stronger focus on the story presented than on surroundings or characters. We labeled this factor as transportation. The remaining items loaded on the second factor which was labeled as identification.

Next, we tested configural invariance by running a MGCFA without cross group equality constraints with transportation and identification and their corresponding items, allowing free correlations among the latent constructs. As per the modification indices, three items were dropped from each latent factor to achieve good model fit (please also see the data analysis documentation in R on OSF). The resulting measurement model demonstrated configural invariance across both samples, χ²(38, N = 556) = 64.53, p = .005, CFI = .99, RMSEA = .050_{[90% CI: 0.028, 0.071]}, SRMR = .028. Factor loadings on the respective latent constructs are displayed in Table 3.

Table 3. Results of the confirmatory factor analysis: factor loadings on transportation and identification.

Factor	Indicator	b	SE	b*	p
Transportation	Id1 “While reading the article, I felt as if I was part of the action”	1.409	0.065	.807	<.001
	Id2 “While reading the article, I forgot myself and was fully absorbed”	1.398	0.068	.776	<.001
	Trans3 “I wanted to learn how the article ended”	1.243	0.074	.671	<.001
	Trans4 “The article affected me emotionally”	1.244	0.068	.713	<.001
Identification	Id5 “I tend to understand the reasons why Robert Hoffmann does what he does”	0.937	0.061	.616	<.001
	Id6 “While reading the article I could feel the emotions Robert Hoffmann portrayed”	1.522	0.066	.837	<.001
	Id7 “During reading, I felt I could really get inside Robert Hoffmann head”	1.332	0.062	.791	<.001
	Id8 “At key moments in the article, I felt I knew exactly what Robert Hoffmann was going through”	1.425	0.065	.810	<.001

With sufficient model fit for configural invariance, we tested metric invariance by fixing the factor loadings to be equal across groups. A change in chi-squared test indicated metric invariance when comparing the fit of the metric model to the configural model, Δχ²(44, N = 556) = 71.811, p = .296. Finally, we tested for scalar invariance by additionally constraining the intercepts to be equal across groups. However, despite good model fit indices, comparing the model fit to the metric model indicated a lack of scalar invariance. Closer inspection revealed that scalar invariance was particularly inflated by the intercept parameters of the items identification2 (“While reading the article, I forgot myself and was fully absorbed”), identification5 (“I tend to understand the reasons why Robert Hoffmann does what he does”), and identification8 (“During reading, I felt I could really get inside Robert Hoffmann head”). We released the parameter constraints by estimating them freely for each country to establish partial measurement invariance. Comparing the revised scalar model to the metric model showed no difference between the models, Δχ²(47, N = 556) = 74.371, p = .465. Thus, partial measurement invariance was established. Some variance was expected due to language differences and the use of a non-validated identification scale, but given the amount of invariance demonstrated, and based on subsequent stratified analyses (not shown) in which we observed the same patterns across cultural contexts, we analyzed the data as one study, as cultural differences did not affect the substantive interpretation of the pattern of results.

Data analysis

To test H₁, we conducted an ANOVA with the message type as the independent variable and message-consistent post-exposure attitudes toward marijuana legalization as the dependent variable. Next, we applied structural equation modeling to test the hypothesized mediation in H₂. H₃ was answered using multiple linear regression and then integrating the findings on the effects of message type and message consistency in another structural equation model.

Results

H₁ predicted that narrative messages would produce stronger message-consistent persuasive effects than non-narrative messages. Although the results were in the hypothesized direction, an ANOVA with message type (narrative, non-narrative) as the independent variable and message-consistent post-exposure attitudes toward marijuana legalization as dependent variable did not provide evidence of a direct effect, F(1, 554) = 0.167, p = .683, η² < .001 (M = 4.24, SD = 2.41 for narratives, and M = 4.16, SD = 2.37 for non-narratives). Thus, H₁ was rejected, as narrative and non-narrative messages did not differentially affect the recipients’ post-exposure attitudes directly.

Next, we applied structural equation modeling based on our invariant measurement model to answer H₂ addressing the indirect effect of message type (X) on post-exposure message consistent attitudes toward marijuana legalization (Y) via transportation (M₁) and identification (M₂). As displayed in Figure 1, the model showed a good fit to the data, χ²(38, N = 556) = 176.217, p < .001, CFI = .95, RMSEA = .081_{[90% CI: 0.069, 0.093]}, SRMR = .045. We found that narrative messages facilitated greater transportation and identification than non-narrative messages. In turn, the greater the level of identification, the greater the recipients’ post-reception issue attitudes. Levels of transportation did not predict recipients’ post-reception issue attitudes. A formal test of indirect effects based on 10,000 bootstrapped samples confirmed that identification mediated the effect of message type on post-exposure issue attitudes. No such indirect effect was observed for transportation. Hence, H₂ was supported for identification but not for transportation. We return to the interpretation of this pattern in the discussion section.

Figure 1. Transportation (M₁) and identification (M₂) as mediators of the effects of the message type (X) on post-exposure message-consistent issue attitudes (Y) in a structural equation model, displayed are standardized regression coefficients.

Notes: N = 556, *p < .05, **p < .01, ***p < .001. Ind. effect Id b* = .085, b = 0.406, 95% CI [0.151, 0.661]; Ind. effect Trans b* = .008, b = 0.038, 95% CI [−0.115, 0.191]; total effect b* = .017, b = 0.083, 95% CI [−0.313, 0.479].

H₃ predicted that the effect of message type on transportation and identification would be moderated by the recipients’ prior attitudes, such that the effect of narrative (vs. non-narrative) messages on transportation and identification would be stronger among respondents exposed to messages that conveyed arguments inconsistent with recipients’ prior attitudes. We formally tested this interaction by conducting a multivariate linear regression with message type, message consistency with the recipients’ prior attitudes, and their interaction as independent variables and transportation and identification, respectively, as dependent variables. In both regression analyses, the predictor variables made a significant explanatory contribution to the recipients’ levels of transportation and identification (see Table 4). However, interactions between message type and consistency were not statistically significant, indicating that prior attitudes did not moderate the effects of message type on transportation or identification. Thus, H₃ was rejected.

Table 4. Multiple regression analysis predicting transportation and identification from message type and message consistency with prior attitudes.

Variable	F	b	SE	b*	t	p	R^2
Transportation	19.52					<.001	.095
Intercept		2.934	0.155		18.872	<.001
Message type		0.148	0.221	.053	0.667	.505
Consistency		0.110	0.034	.188	3.255	.001
Msg type * consistency		0.074	0.047	.141	1.552	.121
Identification	28.73					<.001	.135
Intercept		3.501	0.154		22.711	<.001
Message type		0.527	0.220	.187	2.399	.017
Consistency		0.172	0.034	.289	5.113	<.001
Msg type * consistency		0.011	0.047	.022	0.243	.808

Post hoc analysis

As prior attitudes appeared to be an independent predictor of recipients’ identification and transportation, we included the recipients’ prior attitudes as an additional independent variable (X₂) in our structural equation model with message type (X₁) on post-exposure issue attitudes towards marijuana legalization (Y) via transportation (M₁) and identification (M₂) (see Figure 2). The model showed a good fit to the data, χ²(45, N = 556) = 186.530, p < .001, CFI = .96, RMSEA = .075_{[90% CI: 0.064, 0.087]}, SRMR = .048. Overall, the consistency of a message with the recipients’ prior attitudes emerged as the strongest predictor of the recipients’ post-exposure issue attitudes. The greater the consistency, the greater the post-exposure attitude. In line with the previous mediation model, we found that narrative messages facilitated greater transportation and identification than non-narrative messages. Moreover, results showed that the greater the consistency of the recipients’ prior attitudes with the message they read, the greater their level of transportation and identification. As in the previous mediation model, higher levels of identification were associated with greater persuasive effects, but higher levels of transportation were not. A test of indirect effects based on 10,000 bootstrapped samples confirmed identification as a mediator. No indirect effect was observed for transportation. After accounting for indirect pathways via transportation and identification, there was a significant, negative effect of message type on post-exposure issue attitudes in the overall model. Finally, we found a significant total effect of our model due to the large explanatory power of message consistency with the recipients’ prior attitudes.

Figure 2. Transportation (M₁) and identification (M₂) as mediators of the effects of the message type (X₁) and message consistency with the recipients’ prior attitudes (X₂) on post-exposure message-consistent issue attitudes (Y) in a structural equation model, displayed are standardized regression coefficients.

Notes: N = 556, *p < .05, **p < .01, ***p < .001. Ind. effect Id b* = .064, b = 0.169, 95% CI [0.073, 0.265]; Ind. effect Trans b* = −.023, b = −0.056, 95% CI [−0.121, 0.010]; total effect b* = .938, b = 0.888, 95% CI [0.765, 1.012].

Discussion

This study investigated whether narrative messages generate different levels of transportation and identification than non-narrative messages and have different persuasive effects, when considering prior attitudes toward the issue. We conducted parallel experiments in Germany and the USA to test theory-informed hypotheses about these relationships. Our first set of hypotheses predicted that a narrative message would have stronger persuasive effects on attitudes toward recreational marijuana legalization than a non-narrative message, and that these effects would be mediated by transportation and identification. The narrative message indeed promoted greater transportation and identification than did a non-narrative message. However, multiple regression showed no direct effects of message type in shaping attitudes toward recreational marijuana legalization (although we did find evidence of an indirect pathway between message type and attitudes via identification, a finding that we discuss in greater detail below).

One explanation for the lack of direct message effects on attitudes could be a case of hardened prior attitudes. We tested the effects of prior attitudes using an example that was split in the public opinion toward the issue of recreational marijuana and where, a priori, we had not expected prior attitudes to be hardened or extreme. This was not the case in our samples. Most of our respondents were in favor of recreational marijuana legalization, many held strong views in either direction (“strongly” support or oppose), and most indicated a high level of certainty regarding their position. We thus appear to have a case of strong prior attitudes that may be particularly difficult to change after reception of a single message.

The primary goal of this study was to explore how the recipients’ prior attitudes may shape message processing and effects in the context of narrative and non-narrative messages. Theorists of narrative persuasion have often predicted that narrative messages may be more effective than non-narratives in changing attitudes that are inconsistent with the message’s position (Moyer-Gusé, 2008; Slater & Rouner, 2002). Drawing from this premise and building on the two-step model of defensive processing, we predicted that narrative messages would have a persuasive advantage over non-narrative messages in this context. Furthermore, we reasoned that prior attitudes would moderate the relationship between message type (narrative or non-narrative) and both transportation and identification. However, multiple regression offered no evidence of an interaction between message type and the message consistency in shaping recipients’ level of attitudes, rejecting our hypothesis.

As prior attitudes were confirmed as a strong predictor of the persuasive outcomes, we included message consistency with the recipients’ prior attitudes as an independent variable instead of a moderating variable in our structural equation model. The more the message was in line with the recipients’ prior attitudes, the greater were their levels of transportation and identification with the character. Furthermore, as expected, narrative messages produced greater transportation and identification than non-narrative messages after considering the role of prior attitudes in shaping involvement. Greater levels of identification were, in turn, associated with greater post-exposure message-consistent attitudes towards recreational marijuana legalization. Surprisingly, in light of previous work, we found no evidence of an indirect effect via transportation because transportation was not strongly associated with post-exposure issue attitudes.

This larger pattern of results, and the indirect effect pathways in particular, suggest a potential explanation for the lack of a significant interaction between message consistency with prior attitudes and message type (narrative vs. non-narrative). We unpack the logic behind this interpretation as follows. A key assumption embedded in the logic for expecting greater narrative effects for people with attitudes inconsistent with the position being advocated in a message is that the experience of narrative involvement invites less scrutiny of the arguments and persuasive intent of the narrative message. A key question that emerges from this logic is whether recipients of a narrative message process the narrative with an explicit awareness and scrutiny of the direction of its arguments. In retrospect, it seems likely that different types of narratives offer dissimilar amounts of information about persuasive intent and direction of argumentation. Some forms of narrative (e.g., films, television/streaming shows) may engender a processing state in which recipients do not perceive an obvious persuasive intent or direction. Indeed, many narrative persuasion theorists have assumed a greater potential of narrative over non-narrative messages in the context of fictional and long-form entertainment messages (e.g., Slater & Rouner, 2002), not short-form journalistic storytelling. Other forms of narrative (e.g., advertisements, op-eds, advocacy messages) are likely to make persuasive intent and direction clear from the outset. However, discussions and meta-analyses of narrative persuasion have tended to treat narratives monolithically, without considerations of whether the form of the narrative may lead to different processing (e.g., Braddock & Dillard, 2016; Oschatz & Marker, 2020).

Motivations to defend one’s existing attitude, as outlined in the two-step model of defensive processing, would seem particularly likely under the latter scenarios in which the persuasive intent and direction of argumentation is clear and direct, a prediction consistent with a recently published study (Moyer-Gusé et al., 2019). In other words, if one is motivated to defend their existing attitude, they would seem less likely to become involved in the narrative through narrative transportation or character identification. In turn, this should decrease the likelihood that a narrative would modify recipients’ prior attitudes toward the issue.

The messages tested in the current study indicated a clear position toward the legalization of recreational marijuana immediately – in the title of the article, in an image that accompanied it, and in the opening paragraph. This may have created a reception scenario in which the unique elements of processing a narrative versus a non-narrative message may not have been able to override the elicitation of defensive processing. Such an elicitation may have rendered some narrative processing advantages (in this case, transportation) ineffective in creating a mindset in which respondents were less likely to carefully scrutinize the arguments in the message.

Looking at the pattern of results depicted in Figure 2, the effects of message type (narrative vs. non-narrative) on transportation and identification were either comparable or lesser in magnitude (in terms of standardized coefficients) than the effects of message consistency on these outcomes. This means that any favorable effect of narrative messages on transportation and identification are effectively canceled out by a reduced tendency to become transported and identify with a character when the message content is inconsistent with one’s prior attitude. The absence of a significant relationship between transportation and attitudes toward the topic after reception also suggests that although the narrative stimuli tested in the current study were able to transport respondents more than a non-narrative message, the transportation experience did not translate into a processing state that resulted in greater persuasiveness.

The results do offer some glimmer of hope, however, for narratives embedded within messages that make persuasive intent and directionality of argumentation clear from the outset. Messages of this form that included narrative content were indeed more likely to promote identification with a character, and identification was significantly (albeit weakly) associated with post-reception issue attitudes, producing a significant indirect pathway between narrative message exposure and attitude change. At the same time, the effect of the current narrative on identification was weaker than the effect of message consistency on this outcome, and (as one would expect) prior attitudes remained a very strong direct predictor of post-reception issue attitudes, even after considering transportation and identification. This suggests that it may be possible for counter-attitudinal narratives to create levels of character identification that have potential to modify existing attitudes rather than simply affirming them. However, the narratives used in the current study did not achieve this goal, perhaps owing to their relatively short length, lack of character development, or lack of character and story features that would engender a stronger sense of identification.

We thus offer the (speculative) idea that different narrative formats (or different types of content that surround narrative features) can lead to different reception patterns which affect the likelihood of the narrative preventing or displacing the critical, central processing of a message. In entertainment narrative reception scenarios, a reader might get transported in a narrative message or identify with one character before an inconsistent position, argument, or behavior is presented. In this case, a strong emotional and cognitive connection with the story and character may prevent the processing of the message in a critical systematic way and, thus, facilitate greater persuasive effects. In advocacy/op-ed/advertising narrative reception scenarios where persuasive intent and direction of argumentation are clear from the outset, narratives may struggle to create a transported state of mind that promotes less critical processing of arguments. It may be possible to promote character identification to displace or compete with critical systematic processing, even after a reader is confronted with an inconsistent message, but only in situations where levels of identification are high.

We acknowledge that these are speculative explanations that emerged post hoc and that we do not offer direct comparisons of these processing scenarios in the current study. As noted above, we embedded narrative messages in opinion pieces that had clear and strong issue positions so that we could look at how articles interacted with readers’ preexisting issue attitudes. We chose to hold many elements of these stories constant (e.g., arguments, story format, reasonably short story length) in the interest of isolating narrative effects from arguments or formats. In doing so, however, we did not test narrative messages in entertainment formats despite their popularity in the real world. It thus remains an open question as to whether entertainment and longer-form narratives may interact differently with prior attitudes than the messages studied here. Nevertheless, respondents did perceive the stories with narrative elements to be personal stories, indicating that we did achieve a meaningful degree of narrativity with these stimuli. This approach has also been used in previous studies that have found a persuasive advantage for short-form narrative messages (vs. non-narrative ones), often in the context of news stories (Kim et al., 2012; Oliver et al., 2012; Oschatz et al., 2019), suggesting that such messages can be studied as narratives and offer valuable insight into how narrative messages are processed. It nevertheless remains an open question whether different types and formats of narratives may have differential patterns of effects on narrative processing and persuasion. Testing how audiences process narratives embedded within different message forms would seem an important topic for future study.

A final noteworthy aspect is that we found a partially invariant measurement structure for the latent constructs of transportation and identification across both cultural contexts, meaning that the latent factors comprising the main constructs of interest (transportation and identification) are comparable across both countries. This allowed us to collapse the data into a single analysis and permit stronger inferences that emerge from the observation of similar patterns from two distinct countries with different political systems, languages, and culture. However, the items required some modification to ensure good model fit – moving two items from the identification scale to the measure of transportation and dropping three items from each scale. Future work should explore whether revised measures of transportation and identification reveal similar factor structures in other national, cultural, and language contexts.

Study limitations

This study has limitations that open research potential for future studies. First, although we successfully replicated study results with different populations, we only used a single topic. Future studies should replicate this design with other issues in different fields of research (e.g., health, science communication, or political communication) to see if results are also generalizable across issue topics and messages. Relatedly, this study aimed to test the effects of prior attitudes using an example that was split in the public opinion toward the issue and where no hardened prior attitudes were expected. This was not the case. The majority of our respondents were in favor of recreational marijuana legalization and indicated a high level of certainty regarding their position. We thus appear to have a case of strong prior attitudes that may be unlikely to change after reception of a single message. Further empirical evidence is necessary to investigate the role of strong and/or weak prior attitudes and additional variables that might affect their persuasive impact. We also assume that the clear indication of the article’s position may have prevented respondents from reaching levels of transportation that promote persuasion. A more gentle introduction to the topic may have better chances to overcome critical thinking and change inconsistent prior attitudes, as individuals may have already be transported into the story or have built up a connection to characters (Kaufman & Libby, 2012). We also note that the non-narrative messages we created scored near the midpoint of the manipulation check (personal story/not a personal story), indicating that respondents perceived our non-narrative messages to have some degree of narrativity. We think this is likely due to the fact that we included an identifiable character named Robert Hoffmann, which added a narrative element to the article. This was necessary to compare concepts of transportation and identification across groups and to make the stimulus more realistic as a news article. However, this may have reduced our chance of detecting differences in narrative processing and effects. Finally, we did not observe direct persuasive effects on support for recreational marijuana policy. Our interpretation of results rest on the assumption of a causal effect of transportation and identification based on what is functionally a cross-sectional association between these variables and our measure of attitude, policy support. Although association is a necessary condition for causal inference, we cannot establish temporal order or rule out alternative explanations that may explain this association.

Conclusion

This study makes a contribution to our understanding of the boundary conditions of narrative persuasion. Whereas scholars typically assume that narrative messages are more powerful in changing recipients’ prior attitudes, our results suggest otherwise. Prior attitudes should be considered as an independent influencing factor. In addition, although replicative designs are not common in communication research (McEwan et al., 2018), theory-based psychological models should be examined using different populations (e.g., people from different countries) to test its applicability and generalizability across national boundaries and cultures (Cole, 2003). Future studies should continue this fruitful path to gain confidence in research findings as robust and stable patterns instead of random observations.

Acknowledgements

This research project was developed during a research visit of Dr. Corinna Oschatz in fall 2019 at Cornell University, which was funded by the German Research Foundation (DFG).