Narratives’ Impacts on Attitudes:Do Signaling of Persuasive Intent and Fictionality Matter?

ABSTRACT

Reduced counterarguing – the generation of questions and arguments in response to a message – has been proposed to be a mechanism of persuasion in a variety of contexts, yet many questions remain unanswered regarding the factors that influence this process. Building upon past theorizing in narrative persuasion, this present work investigates whether signaling of persuasive intent (signaling vs. no signaling) and the fictional presentation of texts (fact vs. fiction) decrease counterarguing and, in turn, increase persuasion. Using a 2 × 2 factorial design across four topics at three time points, hypotheses were tested with narratives regarding four controversial political issues, presented either with or without signaling of persuasive intent and in either a news or short fiction format. The online experiment demonstrated that the narratives impacted political attitudes, even when captured in a later follow-up session. However, neither persuasive signaling nor fictional presentation influenced counterarguing or the extent of attitude change, captured both immediately after narrative exposure and again in a follow-up survey two days later.

KEYWORDS:

• Counterarguing
• news
• entertainment
• media effects
• political attitudes
• narrative processing

An ancient Hebrew proverb reads, “The first to present his case seems right, til another comes forward and questions him” (New International Version Bible, 1984, Proverbs 18:7). Investigating what factors prompt audiences to generate questions and arguments against messages, often referred to as counterarguing, is crucial to understanding the process of attitude change. The concept of counterarguing – a sub-component of the larger theory of reactance (Brehm, 1966) – has emerged as a popular area of interest for many scholars, particularly in the area of narrative persuasion. The prevalence of both entertainment and informational narratives in modern society, combined with the extensive time that persons in the developed world spend daily engaging with media messages, has driven a scholarly interest in the attitudinal and behavioral reactions that may be unique to narrative formats. Yet despite extensive literature on the topic of narratives and counterarguing, many questions remain unanswered, particularly in the domain of counterarguing and attitudinal reactance in response to texts focusing on important social issues.

This present work seeks to further narrative counterarguing research by testing previously neglected facets of counterarguing in a social issue context, particularly regarding earlier research which has not clearly manipulated persuasive intent of messages (explicit vs. implicit) in experimental design, but has instead relied on participants’ perceived persuasive intent as a measure. Without this manipulation, it is difficult to support the oft-cited postulation that explicit persuasive intent, in fact, leads to increased counterarguing and reduced persuasion (Moyer-Gusé, 2008; Slater & Rouner, 2002), creating a gap in the literature. Further, past work has theorized that counterarguing in response to a narrative may differ depending on whether the narrative is presented as fictional or factual (Green & Brock, 2000), but little research has empirically tested this relationship. Building upon previous work proposing reduced counterarguing as an explanation for the positive relationship between narratives and attitude change, this discussion and empirical analysis aim to illuminate (a) the impact of upfront signaling of persuasive intent on counterarguing; (b) the relationship between counterarguing and persistence of attitude change in response to implicit vs. explicit persuasive political texts, captured both immediately after narrative exposure and in a follow-up survey two days later; and (c) the way in which persuasion and counterarguing occur in response to fiction vs. nonfiction messages. A justification for this inquiry will be presented in the following pages through a review and discussion of pertinent counterarguing literature, followed by a presentation of hypotheses and an empirical study.

Conceptualizing Counterarguing

The concept of counterarguing finds its roots in the social-psychological concept of reactance. Reactance theory proposes, “When a person is free to make a choice and is then subjected to an attempted usurpation of his choice, he will experience reactance and consequently will tend to reject the influence” (Brehm & Sensenig, 1966, p. 706). Reactance can be conceptualized as a four-step process, beginning with a person first experiencing freedom in some domain, second encountering a threat to that freedom, third reacting to that threat, and fourth experiencing a restoration of the initial freedom (Brehm, 1966; Dillard & Shen, 2005). A threat to freedom in the reactance process need not be physical in order to be considered threatening. For example, a person freely practicing Hinduism may feel intellectually threatened after reading an article arguing that Buddhism is the only truly rational faith. This threat could prompt the Hindu reader to react by internally arguing against the rationality of the article’s claims in order to restore her feeling of confidence in her freely chosen faith.

In the above anecdote, the Hindu reader is engaging in a type of reactance known as counterarguing. Defined as the “generation of thoughts that dispute or are inconsistent with the persuasive argument” (Slater & Rouner, 2002, p. 180), counterarguing is associated with alert thinking, motivation, and careful attention to a message (Futerfas & Nan, 2017; Moyer-Gusé & Nabi, 2010). Although reactance is conceptualized in the literature as including both affective and cognitive responses (Dillard & Shen, 2005; Gardner & Leshner, 2016; Moyer-Gusé & Nabi, 2010), counterarguing is conceptualized as an expressly cognitive response to a threatening message (Dillard & Shen, 2005; Silvia, 2006). As Moyer-Gusé et al. (2011) points out, “Individuals often scrutinize threatening messages, especially those that attempt to change their behavior” (p. 392), emphasizing the fact that a key motivation of counterarguing is perceived threat to freedom of behavior or choice.

Counterarguing and Narrative Persuasion Theory

A wide variety of narrative persuasion research – from advertising to health communication to political communication – has examined how counterarguing influences the effectiveness of persuasive messages (e.g., Escalas, 2004; Green, 2006; Niederdeppe et al., 2011). Counterarguing plays a key role in models and theories explaining the persuasive intent of narratives. For example, the extended elaboration-likelihood model (EELM) of narrative persuasion (Slater & Rouner, 2002) posits that identification with characters and transportation into a narrative world foster narrative persuasion through the reduction of counterarguing. Per the EELM, humans who are deeply engaged with story characters and their experiences (through the process of identification) are not able to step outside the story to generate counterarguments against persuasive messages because they are vicariously living the story through the character. Similarly, people who are absorbed/transported into a narrative world lack the cognitive ability to simultaneously argue against any persuasive subtexts in the story. In essence, mechanisms such as identification and transportation maximize the persuasiveness of narratives through the reduction of counterarguing.

Similar to the EELM, the entertainment overcoming resistance model (EORM) (Moyer-Gusé, 2008) identifies reduced counterarguing as a factor that fosters persuasion. Moyer-Gusé posits that an entertainment message can reduce counterarguing and thus increase persuasion through three different mechanisms: transportation, identification, and parasocial interactions, which are defined as “interaction between an audience member and a media figure such that a pseudorelationship forms” (Moyer-Gusé, 2008, p. 411). Thus, in addition to re-affirming the logic of the EELM regarding transportation and identification, the EORM adds that parasocial interactions can decrease counterarguing because the message source is perceived as trustworthy. Moyer-Gusé (2008) suggests that “trust of and familiarity with characters that viewers develop can have important effects on willingness to accept information, even persuasive, without counterarguing with the source’s claims” (p. 416).

The unique ability of narratives to reduce counterarguing has been supported by several research endeavors which have compared narratives to non-narrative-based arguments in varied contexts. Deighton et al. (1989) compared argument-based television commercials to drama-based television commercials, distinguished from argument by distinct protagonists and a story plot. The authors found that the drama-based commercials resulted in significantly less counterarguing than the argument-based commercials. In the context of health communication, Gardner and Leshner (2016) compared non-narrative informational messages with narratives featuring an exemplar telling his/her story in first-person and found that the narrative message resulted in significantly less counterarguing than the purely informational message. Further, Kreuter et al. (2010) compared a breast cancer video featuring stories from survivors to a non-narrative informational video and found that the video featuring stories led to less counterarguing and more persuasive effects than the informational video.

Perceived Persuasive Intent and Signaling

The persuasive power of narrative is thought by some to be found in narratives’ implicit approach to persuasion. Communication scholars have long theorized that perceived persuasive intent increases a person’s reactance and motivation to counterargue. Persuasion attempts can be interpreted by humans as threats against their freedom. Thus, scholars argue that humans feel more threatened by explicit persuasive arguments than by implicit persuasive arguments (Green & Clark, 2013). It follows, then, that since counterarguing is a type of cognitive reactance to threat, counterarguing should occur in response to explicit persuasive messages more than implicit persuasive messages. “Explicit” persuasive arguments often include persuasive signaling – some type of upfront indicator that the message to come is intended to be persuasive, such as a provocative title or the labeling of an opinion-editorial piece in a newspaper. In contrast, entertainment narratives that include subtle persuasive elements but no clear persuasive signaling would be considered “implicit” persuasive arguments, a feature which is thought to aid their persuasiveness (Moyer-Gusé, 2008). For example, in narrative research on smoking messages, Green and Clark (2013) argued that explicit vs. implicit persuasive messages can significantly impact counterarguing and persuasive effectiveness, writing, “Individuals may not counterargue implicitly presented messages (such as a brief positive representation of smoking) because they may not realize they are being persuaded” (p. 480). Yet, thus far in the research, the relationship between explicit persuasive intent and decreased persuasion through counterarguing has not been robustly established. In health contexts in particular, a review of narrative persuasion research led de Graaf et al. (2016) to the conclusion that “an overtly persuasive context does not necessarily preclude narrative effects in a health context (p. 89).”

This idea that perceived persuasive intent should lead to increased counterarguing has found some support in empirical research, but evidence without confounds is lacking. Moyer-Gusé and Nabi (2010) found in their study regarding teen pregnancy narratives (discussed earlier) that perceived persuasive intent was positively correlated with increased cognitive reactance. However, persuasive signaling was not manipulated in the design; rather, persuasive intent was simply measured as a response to the narrative and non-narrative messages presented in the study, leaving open the possibility that persuasive intent was confounded with the potentially differing persuasive strength of the messages. Further, Dillard and Shen (2005) found in a health communication study that more threatening language in a persuasive message led to more perceived intent to persuade and cognitive reactance (a concept similar to counterarguing). However, the fact that the authors confounded persuasive intent with language intensity raises the question of whether the perception of intent or the language intensity itself caused the increase in reactance.

Overall, then, current narrative persuasion theorizing makes a strong case for counterarguing as a mechanism of narrative exposure and persuasive outcomes. Further, explicitness of the signaling of a message’s persuasive intent is conceptualized as a message attribute that is likely to predict how much counterarguing occurs in response to a narrative. However, non-confounded tests of this relationship are lacking in the literature. In order to further the understanding of the process of narrative persuasion, we next present four hypotheses focusing on the persuasive impact of narratives and the predicted effects of persuasive signaling on counterarguing and persuasion.

First, at a foundational level, the wide body of literature reviewed thus far is situated in the well-supported claim that narratives hold persuasive power in a variety of contexts – including political and social issues. Narratives can possess a unique ability to overcome reactance to persuasive messages by presenting worlds and characters that promote deep engagement (Moyer-Gusé, 2008; Slater & Rouner, 2002). Further, in examining the persuasive power of narratives, the current study measures persistent attitude change (i.e., enduring effects that last several days) in line with research that calls for more studies to incorporate follow-up surveys to determine the lasting impact of narratives (e.g., Knobloch-Westerwick et al., 2020; Perrier & Martin Ginis, 2018). This fundamental logic of narrative persuasion theorizing leads us to our first hypothesis:

H1: Exposure to a narrative text regarding a political topic has (a) immediate and (b) persistent positive impact on text-consistent attitudes regarding the political issue covered in the text.

Further, as presented in our literature review, current research lacks a clear, non-confounded test of the oft-cited idea that explicit persuasive intent increases counterarguing, which impedes persuasion (Green & Clark, 2013; Moyer-Gusé, 2008; Slater & Rouner, 2002). Despite the strong theoretical arguments made regarding the effect of explicit persuasive intent on counterarguing, prior research has not isolated persuasive intent from other potential factors contributing to counterarguing, leading to a gap in the literature. A careful, non-confounded investigation into this claim at the heart of counterarguing and persuasion theorizing is needed. Thus, the following three hypotheses are proposed for testing in consistency with current narrative persuasion models, including the EELM and the EORM (Green & Clark, 2013; Moyer-Gusé, 2008; Slater & Rouner, 2002):

H2: Cues signaling persuasive intent of the text increase counterarguing.

H3: Cues signaling persuasive intent of the text reduce text-consistent attitude change as compared to texts with no persuasive signaling cues.

H4: The effect postulated in H3 is mediated by counterarguing.

Fiction and Narrative Persuasion

In addition to these four hypotheses addressing narrative persuasion and persuasive signaling, we present a brief review of an additional factor considered relevant to narrative persuasion, along with a fifth hypothesis and one research question. An important issue discussed in counterarguing literature is whether factual or fictional messages might produce more counterarguing. Following the logic of persuasive intent, people may be more vulnerable to persuasion (and less likely to counterargue) when exposed to what they believe to be fiction, because fiction is often viewed as entertainment, which may not be perceived to be created for the intent of persuasion. Yet studies that have investigated potential persuasive differences between factual and fictional narratives suggest that both are about equally persuasive (e.g., Appel & Richter, 2007; Butler et al., 1995; Green & Brock, 2000). Given this research, as well as recent findings supporting the idea that political narratives presented as fact vs. fiction are equally persuasive (Knobloch-Westerwick et al., 2020), the following hypothesis is proposed as a replication of the above-cited previous work:

H5: Factual accounts (presented as news) and fictional accounts (presented as short “flash fiction”) will result in equivalent (a) immediate impact and (b) persistent impact on attitudes.

Digging deeper than the direct persuasive effects of fact vs. fiction, very few researchers have empirically tested the potential relationship between factual vs. fictional presentations and counterarguing. Cao (2015) set out to test this relationship using short video documentaries that could be reasonably interpreted as either fact or fiction depending on their label. Cao found that participants who had a low need for cognition counterargued against fiction significantly less than they counterargued against fact, whereas there was no difference in counterarguing among participants high in need for cognition. Given the general dearth of research in this area, the following research question is offered in order to better understand what relationship between factual vs. fictional messages and counterarguing exists in a broader social context:

RQ1: If a text presents fictional (as opposed to factual) events, is counterarguing reduced?

Method

Overview

Adult participants (N = 171) completed a two-session study online as approved by the university’s Institutional Review Board. The experimental (between-subjects) design varied two distinct text features: format (fact vs. fiction) and persuasive intent (signal vs. no signal). In order to vary format, stimuli were situated in either a fiction book style template or a newspaper template with very minimal edits to the text made to maintain stylistic consistency with fiction vs. news genres. Further, to vary persuasive intent, clear signals were added to the templates for both the fact and fiction versions. The experimental design also included two within-subjects factors: text topics (immigration, abortion, healthcare, and death penalty) and time (t1 attitudes measured before text exposure, t2 measured after exposure, and a t3 delayed measure of attitudes completed two days after exposure). Thus, overall, the experiment featured a 2 × 2 factorial design across four topics at three time points.

Participants

Although 196 participants completed session one, 15 of them were excluded because they spent either less than 10 minutes or more than four hours on the entire session, which suggests they did not allocate sufficient, uninterrupted attention to the survey. Additionally, consistent with prior research using similar stimuli (Knobloch-Westerwick et al., 2020), 10 participants were excluded because their reading time on the stimulus text pages was either extremely short or long, below 20 or over 600 seconds,¹ indicating a lack of full engagement with the texts. Analyses were thus based on 171 participants, with experimental cells ranging between 40 and 45 participants. The sample consisted of 100 women, 70 men, and one person who did not indicate gender information. The average age was 20.74 years (SD = 3.33). One third (33.3%) identified as Democrat, 27.5% identified as Republican, 30.4% categorized themselves as independent, and 8.8% chose none of these categories. For the follow-up survey (t3), the sample size was smaller (N = 116) due to attrition.

Procedure

Recruitment

Participants were adult college students recruited via e-mail through undergraduate journalism and communication classes and through an undergraduate participant pool at a large public university. Participants received course credit as an incentive. All elements of the experiment were completed by the participants online.

Baseline Measures (T1)

After consenting to the research, participants completed attitude measures regarding the four target issues (immigration, abortion, healthcare, and death penalty) and several distractor issues. Additionally, participants reported their perceived importance of these issues. Participants also reported the time they spent engaging with news the day before.

Stimuli Exposure and Text-specific Measures

After finishing the baseline measures (t1), participants were informed that they had advanced to the second section of the first session and were asked to carefully read four texts. Participants were randomly assigned to one of four conditions varying fact vs. fiction and signal vs. no signal: news format with persuasive signaling, news format without persuasive signaling, short fiction format with persuasive signaling, and short fiction format without persuasive signaling. Each participant read four texts (presented one at a time) of different topics with the same format and persuasive intent signaling. The order in which participants read texts on the four different topics (immigration, abortion, healthcare, and death penalty) was randomized across participants. No time limits were given, and the survey software recorded how long each participant spent on each story. After each story, participants answered questions measuring counterarguing, and narrative evaluations in response to each individual text.

Postexposure Measures (T2)

After completing the readings and text-specific questions, participants began the third section of the first session. Participants responded to the same issue attitude and importance questions asked in the baseline measures (t1). Participants then began the fourth section where they answered manipulation checks regarding the genre of texts they read, perceptions of fact vs. fiction, perceptions of political stance, and perceptions of narrative intent. Participants also answered questions that measured their need for cognition and reported their political affiliation and demographic information. Participants were then thanked and informed that they would receive instructions for completing session two in two days.

Follow-up Session (T3)

Two days after completing the first session, participants were emailed a link to a short follow-up survey. The survey included questions from the baseline (t1) and postexposure (t2) measures regarding political attitudes and perceived importance of political issues. Further, participants completed a manipulation check regarding the perceived persuasive intent of the text and again reported their time spent consuming news the day before. Finally, participants were thanked and debriefed.

Stimuli

Content

Persuasive political texts were adapted from the authors’ previous work in political persuasion (Knobloch-Westerwick et al., 2020). Four controversial social topics were used as the subject for the various texts: immigration, abortion, healthcare, and the death penalty. For each of the four topics, four text versions were created (news format with persuasive signaling, news format without persuasive signaling, fiction format with persuasive signaling, fiction format without persuasive signaling), resulting in a total of 16 text stimuli. Each text was exactly 600 words long with a five-word title and a byline featuring an author name that was gender-nonspecific. Each text was loosely patterned after real news stories and featured clear protagonists and extensive direct quotations/dialogue throughout. For each political topic, all four versions of the text were very similar and featured the same clear political stance, with the abortion and healthcare topics indicating a conservative position and the immigration and death penalty topics indicating a liberal position. However, none of the texts made any explicit statement regarding a stance on the issue. Manipulation checks (reported below under “political stance” in the results section) affirmed that the implied political stances were clearly understood by participants.

Format

Texts were presented in both news and short fiction formats to fulfill the fact vs. fiction manipulation. In order to achieve a realistic manipulation, two distinct text displays were developed by a graduate student with professional publication design experience. For the news display, an online template was modeled off of several major news websites and featured the label, The Star Gazette. For the short fiction texts, the display was designed to appear as an online book reader and featured a large thumbnail of the book cover that describes the book as fiction (see Figure 1a-1c in the online supplement for examples²). Manipulation checks (reported below under “format” in the results section) affirmed that the fact vs. fiction manipulation was successful.

Persuasive Intent

In addition to varying in their presentation as either news or fiction, the texts also varied in whether or not they indicated persuasive intent (signal vs. no signal). For the news format, persuasive signaling was accomplished through the addition of the words “opinion editorial” in the signal condition at the top of the article, in large red font and all capital letters. Additionally, the word “opinion” was added to the byline. For the fiction format, persuasive signaling was differentiated through the title of the book featured on the thumbnail cover and page headers. In the signal condition, the book title was The Advocate: Short Fiction with a Stance, with “stance” featured in large, red capital letters. In the no signal condition, the book title was Imagined World: A Collection of Fictional Stories (see Figure 1a-1c in the online supplement for examples). Manipulation checks (reported below under “persuasive intent” in the results section) affirmed that the signal vs. no signal manipulation was successful.

Measures

Attitudes

Embedded in six distractor statements, the four prompts (reported in the online supplement under “Appendix”) served to capture participants’ attitudes regarding the four target issues that were also the topics of the stimuli texts. Each target issue was measured with one prompt where participants indicated their attitudes on a slider scale recording measurements of −50 to 50, with strongly oppose and strongly support as anchors. Appendix A reports descriptive statistics. Note that the third measurement point was completed by fewer participants, due to attrition (details reported in Appendix A).

For the analyses testing persuasive impacts, the attitude measures were coded such that higher scores indicated greater agreement with the text stance (e.g., the attitude scores that originally indicated how much a participant supported the death penalty were reverse-coded, because the text favored abolishing the death penalty). In addition, for statistical analyses, immediate attitude impacts were calculated by subtracting t2 attitude scores from t1 scores. Persistent attitude change was calculated by subtracting t1 attitude scores from t3 scores in line with previous narrative persuasion research (Knobloch-Westerwick et al., 2020).

Counterarguing

Counterarguing was measured with eight items adopted from Moyer-Gusé et al. (2018), such as “While I read, I felt skeptical of the position the text seemed to be advocating.” Three items were reverse coded. The reliability per Cronbach’s alpha ranged from .80 to .89 for the four texts and was .87 across all stories. Based on satisfactory inter-item consistency, an overall counterarguing score was created based on an average across all measures, as levels of counterarguing did not differ drastically by text, with means ranging between M = 3.24 (SD = .99) to M = 3.86 (SD = 1.38). The overall counterarguing variable had a mean of M = 3.48 (SD = .72). Counterarguing was generally a covariate in statistical analyses.

Exposure

The online survey procedure captured the amount of time that participants spent on the text pages. The average seconds participants spent viewing a single text was M = 154.15 (SD = 84.46).

Manipulation Checks

The following manipulation checks were assessed. Results for these manipulation checks are reported in the results section.

Signaling of Persuasive Intent

At the end of the follow-up session (t3), participants were shown a screenshot of the front page of an example of each text format (fictional with persuasive signaling, fictional without persuasive signaling, factual with persuasive signaling, and factual without persuasive signaling) accompanied by the prompt, “Does the presentation below signal that the text aims to PERSUADE?” and two distractor items. Answers to the prompt were measured on a seven-point Likert-type scale, ranging from strongly disagree to strongly agree.

Format

Perceptions of the fact vs. fiction (FF) format of the texts were assessed with two manipulation checks, both during the postexposure measures (t2). For one manipulation check, participants answered the prompt “Did the described events really happen and were the described individuals real people? Or were the described events and people all fictional? Please indicate your impression for each of the texts.” Answers to this prompt were recorded for each text using a slider scale recording measurements from −50 to 50, with Real Events/People” and Fictional Events/People as anchors.

A second manipulation check asked participants the following prompt: “Which type of texts did you read? Please indicate which text type applies.” Participants were then presented with five slider scales, labeled “News Reports,” “Short Fiction Stories (‘flash fiction’),” and three distractor labels. The slider recorded measurements ranging from −5 to 5, with anchors labeled Not applicable at all and Definitely applicable.

Political Stance

Participants encountered a question for each text asking whether the text supported or opposed the issue in the text (i.e., “Does the text about Leonard Curtis (‘A Struggle on Death Row’) oppose or support the death penalty?”) Participants responded to each question with a slider which recorded measurements ranging from −50 to 50, with anchors labeled Strongly OPPOSES [the issue] and Strongly SUPPORTS [the issue].

Additional Measures

Participants also completed a series of self-report items measuring elaboration, need for cognition, recent news consumption, narrative evaluation, perceived narrative intent, political issue importance, and political participation; however, these variables were not used in the analyses for this study.

Results

Manipulation Checks

Signaling of Persuasive Intent

The measures capturing participants’ impressions of the four different presentation formats – fictional with persuasive signaling, fictional without persuasive signaling, factual with persuasive signaling, and factual without persuasive signaling – were subjected to an ANOVA with four repeated measures. The fact vs. fiction (FF) and persuasive intent (PI) presentations were within-subjects factors in this case because the follow-up session presented all four display types to each participant. The analysis showed significant effects for PI, yielding F(1, 97³) = 8.94, p = .004, η2 = .019 Further, the analysis showed an effect approaching significance for FF, yielding F(1, 97) = 3.90, p = .051, η2 = .009. The presentation as factual produced higher ratings for perceived persuasive intent, M = 5.60 (SD = 1.13), than the fictional presentation, M = 5.37 (SD = 1.28). The presentation with persuasive signaling led to higher ratings for perceived persuasive intent, M = 5.65 (SD = 1.22) vs. M = 5.32 (SD = 1.17), corroborating that the persuasive intent manipulation was successful. A correlation matrix featuring key variables is available in Appendix B.

Format

The measures capturing whether the presented events were perceived to be factual vs. fictional (fictionality) for each story were subjected to an ANOVA with four repeated measures for the four texts, differentiated by the between-subjects factors of fact vs. fiction (FF) and persuasive intent (PI). Only FF showed a significant effect, yielding F(1, 167) = 17.02, p < .001, η2 = .049. The presentation as factual produced lower ratings for fictionality, M = −1.21 (SD = 2.09) than the fictional presentation, M = .42 (SD = 3.03), corroborating that the factual vs. fictional manipulation was successful.

Another manipulation check was based on ratings on whether participants had read news reports or short stories. An ANOVA with these two dependent repeated measures and FF and PI as between-subjects factors yielded that the manipulation was successful, F(1, 166) = 60.71, p < .001, η2 = .202, because participants who read the fiction versions scored high on the rating for short fiction stories, M = 79.70 (SD = 27.37), and low on the rating for news report, M = 33.49 (SD = 30.37), while participants who read the news versions scored lower on the ratings for short fiction stories, M = 56.58 (SD = 36.17), and high on the ratings for news reports, M = 71.71 (SD = 27.76).

Political Stance

Based on the questions on political stance of the texts, all four texts, in fact, conveyed a clear political stance as desired, because the ratings differed significantly from the scale midpoint, p < .001 for all stories, with differences from scale-midpoint ranging between M = 5.53 (SE = 2.54) and M = 17.0 (SE = 1.80).

Immediate Attitude Impacts (H1a, H3, H5a)

To test H1a regarding immediate attitude impacts from exposure to a text regarding a political topic, an ANOVA used the attitudes scores, measured before (t1) and right after (t2) text exposure, as a within-subjects factor for time. The four texts served as an additional within-subjects factor. Furthermore, the experimental manipulations of persuasive intent (PI) and fact vs. fiction (FF) were incorporated as between-subjects factors. Additionally, given that narrative persuasion literature proposes that counterarguing inhibits persuasion (Moyer-Gusé, 2008; Slater & Rouner, 2002), the analysis included counterarguing as a covariate.

The within-subjects factor of time had a significant impact, F(1, 158) = 11.14, p = .001, η2 = .011, because attitudes shifted in line with the texts’ stances, based on the before/after comparison. At t1, the average attitude score across topics was M = −6.04 (SD = 13.04); at t2, it was M = .18 (SD = 13.43). The mean difference for the pre-post test scores (based on t2-t1) was M = 6.21 (SD = 10.34). This finding supports H1a. As no significant interactions were observed with the within-subjects factor for texts, this effect was robust across topics. The covariate of counterarguing, F(1, 158) = 3.23, p = .074, η2 = .003, approached significance, with greater counterarguing reducing attitude change, r = −.17 (p = .026). In contrast to H3, persuasive intent (PI) also had no effect on attitude change (p = .323, η2 = .001). Further, in support of H5a, fact vs. fiction (FF) had no effect on attitude change (p = .612, η2 < .001). No additional relevant effects approached significance in this analysis.

Considering the similarity in means between the fact and fiction conditions, as well as the prediction of equivalence of the effect of fact vs. fiction (FF) on attitude change in H5, we assessed evidence of statistical equivalence. In an independent-samples equivalence procedure (see Weber & Popova, 2012), the fact vs. fiction (FF) manipulation served as the grouping variable and the mean change score for attitudes (t2-t1) as the dependent variable. Due to the general lack of prior experimental research on this relationship (and the corresponding lack of previously observed effect sizes), equivalence was tested at a range of effect size values proposed by Cohen (1988) as broad guidelines for the social sciences. Evidence was found for equivalence assuming a large (delta = .50), medium (delta = .30), and medium-small (delta = .20) effect size (t(167) = −.34, ∆ = .50, p_eq < .001; t(167) = −.34, ∆ = .30; p_eq = .002; t(167) = −.34, ∆ = .20, p_eq =.039). Evidence for equivalence reaches nonsignificance at ∆ < .19. Thus, overall, the evidence is generally consistent with statistical equivalence.

To examine whether the lack of support for H3 stemmed from a lack of statistical power, a power analysis was run based on the effect size of findings in past research measuring the impact of perceived persuasive intent on attitude change. Although the small number of studies reporting the effect of persuasive intent on attitudes, as well as the inherent confounds in persuasive intent manipulations in the literature up until this point (discussed above), make these comparisons imperfect, this analysis can provide an estimate of the power needed to detect effects of a relevant size. As described earlier in this article, Moyer-Gusé and Nabi (2010) found that perceived persuasive intent of a dramatic narrative had a significant effect on attitudes and intentions. The authors report a significant correlation with a small/medium (per Cohen, 1988) effect size (r = .18) between perceived persuasive intent and perceived vulnerability, an attitudinal measure. Further, Silvia (2006) studied reactance using a form of up-front signaling by varying the presence or absence of coercive statements in persuasive messages to manipulate people’s perceptions of the message and elicit reactance. Silvia’s manipulation had a medium-sized significant effect on message-consistent attitudes, such that messages with a coercive statement near the beginning were significantly less persuasive than messages without the coercive statement (t(78) = 3.47, p <.001, d = 0.77). Given this precedent of a small/medium to medium effect size of perceived persuasive intent on attitudinal outcomes, a power analysis using GPower (Erdfelder et al., 1996) was conducted, with power⁴ (1 – β) set at 0.80 and α = .05, two-tailed and a conservative, small/medium effect size (as defined by Cohen, 1988) of f = .15. The resulting analysis showed that a total sample size of 128 would be needed to detect an effect of this size. Thus, considering that our sample size far exceeds this number (N = 171), it is unlikely that the lack of support for H3 resulted from insufficient sample size; instead the postulated effect appears too small to be practically relevant.

Persistent Attitude Impacts (H1b and H5b)

To test H1b’s prediction of persistent attitude impacts, an ANOVA used the attitude measures assessed before the text exposure (t1), immediately after (t2) exposure, and two days later (t3) as repeated measures. Once more, the four texts served as an additional within-subjects factor. Again, the experimental manipulations of persuasive intent (PI) and fact vs. fiction (FF) were incorporated as between-subjects factors, while the analysis included counterarguing as a covariate. Note that the sample size for this analysis was smaller (n = 116) due to attrition at the third attitude measurement point.

The within-subjects factor of time had a significant impact, F(2, 206) = 7.72, p = .001, η2 = .012, because attitudes shifted in line with the texts’ stances, based on comparisons across time. Per the estimated marginal means, at t1, the average attitude across topics was M = −4.32 (SE = 1.14). At t2, the mean attitude score was M = 1.23 (SE = 1.27), which was significantly different from t1 (p < .001, pairwise comparison with Sidak correction). At t3, the mean attitude score was M = −.732 (SE = 1.25), which was significantly different from both t1 (p < .001) and t2 (p = .033). The mean differences for the attitude scores (based on t2-t1) was M = 5.55 (SE = .95) and (based on t3-t1) M = 3.59 (SE = .89), while the immediate and the delayed measures (t3-t2) differed by M = −1.96 (SE = .76). In other words, the delayed measure showed that the impact of the text exposure diminished slightly over time but was still significant. This finding supports H1b. Since there were no interactions with the within-subjects factor for texts, this effect was robust across topics.

The covariate of counterarguing, F(2, 206) = 3.62, p = .028, η2 = .006, was significant: Greater counterarguing reduced immediate attitude change per t2-t1, r = −.23 (p = .015). Further, looking at persistent attitude change per t3-t1, counterarguing approached a significant influence, r = −.17 (p = .064).

In this analysis, the findings did not support H3, as PI also had no effect on attitude change (p = .578, η2 = .001).⁵ No additional relevant effects approached significance in this analysis. Further, as predicted in H5b, the FF manipulation once more did not significantly affect attitude change (p = .876, η2 < .001).

Additionally, due to the prediction of equivalence in H5b, we assessed evidence of statistical equivalence using the same procedure as described for H5a (see Weber & Popova, 2012), except in this instance, the mean change in attitude scores from t1 to t3 was used as the dependent variable (t3-t1). We find evidence for equivalence assuming both large (∆ = .50) and medium (∆ = .30) effect sizes (t(112) = −.55, ∆ = .50, p_eq < .001; t(112) = −.55, ∆ = .30, p_eq = .018). Evidence for equivalence reaches nonsignificance at ∆ < .26. Thus, we observe the evidence to be generally consistent with statistical equivalence.

Impacts of Persuasive Intent and Fact Vs. Fiction on Counterarguing (H2, RQ1)

The postulations that both signaling of persuasive intent (per H2) and factual texts (compared to fictional texts, per RQ1) would increase counterarguing were tested with an ANOVA with PI and FF as between-subjects factors. None of the effects approached significance in this analysis (p = .395, η2 < .001 for PI; p = .984, η2 = .000 for FF). Thus, the postulations in H2 and RQ1 were not supported.

Influence of Counterarguing as Mediator of Attitudinal Impacts (H4)

A mediation analysis (Model 4) using PROCESS SPSS version 3.2.01 (Hayes, 2013) tested predictions per H4 that cues signaling persuasive intent of the text reduce attitude impacts via counterarguing. A point estimate for an indirect effect (total or specific) was considered significant if zero was not included in the 95% bias-corrected confidence interval. For the mediation analysis, signaling of persuasive intent served as the independent variable (X, 0 = no persuasive signaling, 1 = persuasive signaling), counterarguing was the mediator (M), and the dependent variable was mean differences for the immediate attitudinal impacts (based on t2-t1) (Y). The mediation produced no significant results (ns); thus, H4 was not supported. A second mediation analysis examined mean differences for persistent attitude impacts (t3-t1) and was not significant (ns).

Discussion

Narrative persuasion literature has not provided a clean empirical test of the theoretical assertion that signaling of explicit persuasive intent increases counterarguing and reduces persuasion (Green & Clark, 2013; Moyer-Gusé, 2008; Slater & Rouner, 2002). Through manipulating persuasive intent through explicit persuasive signaling in a carefully controlled experimental design, this study tested whether explicitness of persuasive intent alone (without various confounds) increases counterarguing. The fact that the successful persuasive intent manipulation yielded no significant effect on counterarguing or persuasion across four distinct texts/topics calls into question the current theoretical understanding of the effect of implicit vs. explicit persuasive intent on counterarguing and persuasion.

H1 predicted that exposure to a text regarding a political topic would influence (a) immediate and (b) persistent impacts on attitudes pertaining to the political issue in the text. H1a was supported, indicating that participants’ attitudes shifted in line with the texts’ stance immediately after message exposure. Further, H1b was also supported, demonstrating a lasting impact of the texts two days after exposure, with participants’ political attitudes still in line with those portrayed in the texts, although the effects diminished somewhat. The fact that participants experienced a significant shift in attitudes across four politically-controversial topics after reading a 600-word text for each topic demonstrates the power of narrative persuasion in a political context. Further, the fact that significant persuasive effects persisted even 48 hours after participants read the texts reiterates the lasting power of narrative.

Additionally, H5, which predicted that accounts presented as factual vs. fictional would not differ in their (a) immediate impact and (b) persistent impact on attitudes, was supported, lending further support to the idea that narratives can be persuasive regarding real-world political issues regardless of whether the narratives themselves are true stories. Although the idea that stories known to be fictional can still persuade audiences on real-world issues may seem counterintuitive, limited research has suggested that fiction may be uniquely powerful in reducing counterarguing, thus enhancing persuasion (Cao, 2015). Yet other work has suggested that narrative structure reduces counterarguing and increases persuasion regardless of factual vs. fictional presentation (Green & Brock, 2000; Slater & Rouner, 2002). In an effort to further explore this relationship, the research question (RQ1) asked if the presentation of a text as fictional (as opposed to factual) reduced counterarguing. No significant difference was found in counterarguing in response to factual vs. fictional texts. Thus, this finding calls into question the proposition that decreased counterarguing could function as a consistent mechanism for the persuasive effects of fiction. Given the mixed results of research studies on this topic (discussed above), further research should investigate whether fiction as a mechanism for decreased counterarguing may be a context-specific phenomenon, examining how humans may approach different types/genres of fiction with varying expectations.

Taken together, hypotheses 2–4 addressed the relationship between persuasive intent, counterarguing, and attitude change, testing the predominate view of the effects of implicit vs. explicit persuasive intent in narrative persuasion theory (Green & Clark, 2013; Moyer-Gusé, 2008; Slater & Rouner, 2002). H2 predicted that cues signaling persuasive intent of the text would increase counterarguing. Further, H3 predicted that cues signaling persuasive intent of the text would reduce attitude impact. Finally, H4 predicted that counterarguing would mediate the relationship between persuasive signaling and attitude impact. For the manipulation of persuasive intent, identical texts were used in each condition, but they were placed in templates with or without persuasive signaling. Despite the fact that manipulation checks supported the success of this manipulation, no significant difference emerged in counterarguing or attitude impact between texts with or without persuasive signaling. Further, there was no evidence that the mediation predicted in H4 occurred. Thus, contrary to prior theorizing in narrative persuasion, H2, H3, and H4 were not supported.

The lack of difference in counterarguing and attitude change between those texts which demonstrated explicit persuasive intent (through signaling) and those that did not poses an interesting question, particularly since a power analysis indicated that the lack of effect of persuasive signaling on attitudes did not result from insufficient power. Is perceived persuasive intent itself a predictor of counterarguing and attitude change, or do actual message differences drive the relationship? As discussed earlier in this article, current literature lacks non-confounded tests of the effects of perceived persuasive intent on counterarguing and persuasive outcomes. Prior research attempting to test the impacts of explicit vs. implicit persuasive intent on counterarguing and persuasion has featured potential confounds, including narrative vs. non-narrative manipulations and variance in language intensity of the persuasive content (Dillard & Shen, 2005; Moyer-Gusé & Nabi, 2010). Thus, past research has measured perceived persuasive intent in response to texts or videos with fundamentally different content, not identical texts with differences of signaling. This study’s findings suggest that humans’ perception of persuasive intent may not inherently drive an increase or decrease in counterarguing; rather, the message features that typically accompany persuasive intent (i.e., intense language and argument structure) could be triggering counterarguing and impeding persuasion. Further research in broader contexts testing explicit vs. implicit persuasive intent without confounds in message content would strengthen theoretical understanding of the process of narrative persuasion.

Although instructive, this study does feature several limitations. First, the use of an undergraduate student sample is not representative of the full diversity of the U.S. population. Second, although the manipulation of persuasive signaling was found to result in statistically significant differences in perceived persuasive intent, the mean difference in perceived signaling of persuasive intent was small (0.33 on a 7-point scale). Further, the fact that participants rated perceived signaling of persuasive intent well above the midpoint for both conditions (5.65 for signaling condition; 5.32 for no signaling condition) indicates a strong perception of persuasive intent was present across the board, though significantly more strongly in the signaling condition. Thus, the possibility remains that the signaling manipulation was not dramatic enough to cause readers to sense a comparatively strong threat to their freedom and thus failed to initiate the psychological reactance process that should lead to counterarguing. Although the design sought to mimic real-world examples of persuasive signaling and featured objective differences in presentation (providing experimental value even apart from manipulation checks – see O’Keefe (2003)), it is difficult to predict just how explicit message signals might need to be in order to cause readers to feel threatened. Further research could devise alternate manipulations and probe varying levels of intensity of persuasive signaling and its impact on reactance and counterarguing. Finally, the unexpected findings regarding persuasive intent spark the question of whether persuasive intent could operate differently in political contexts than in other contexts. For instance, it is possible that since politics is an inherently contentious world, readers may expect a certain level of persuasive intent in any political narrative and may be more tolerant of persuasive messages in political contexts than they would be in other contexts. Overall, this study lends new insight to narrative persuasion in a political/social issue context and highlights the need for more rigorous testing and exploration of key concepts in narrative persuasion theory.

Acknowledgments

The authors acknowledge the three anonymous reviewers of this article for their careful and helpful reviews. Further, the authors acknowledge Dr. Matthew Grizzard and Dr. Emily Moyer-Gusé for their helpful insights in the refinement of this work.

Disclosure Statement

No potential conflict of interest was reported by the authors.

Additional information

Notes on contributors

Rebecca Frazer

Rebecca Frazer (MPA, The Ohio State University, 2019) is a PhD student in the School of Communication at The Ohio State University. Her research focuses on media narrative processing and effects, with an emphasis on moral and political judgments.

Melissa J. Robinson

Melissa J. Robinson (PhD, The Ohio State University, 2017) is an assistant professor at Penn State Fayette, The Eberly Campus. She addresses important questions regarding the psychological and behavioral effects of mediated health communication on the individual, as well as the underlying persuasive processes occurring.

Silvia Knobloch-Westerwick

Silvia Knobloch-Westerwick (PhD, Hanover University of Music, Drama, and Media, 1999) is a professor in the School of Communication at The Ohio State University and serves as coeditor of Communication Research. Her research interests include the selection, processing, and effects of mediated communication. A key thread in her work pertains to antecedents and consequences of selective exposure to mediated messages.

Notes

1. Prior to this exclusion (N = 186), the mean exposure time for participants was 158.11 seconds (SD = 172.40). The nine participants excluded from analysis for spending < 20 seconds on average on each text constituted the lowest 5.0% of participants, while the participant excluded for spending > 600 seconds on average on each text spent an average of 2144.07 seconds on each text, which was more than four times longer than the next longest-reading participant, identifying this participant as a clear outlier.

2. The online supplement can be accessed at https://osf.io/6ka83/

3. The n for the PI manipulation check (n = 98) is slightly smaller than the n for the other t3 measures (n = 116) because the measure was not included in the survey distributed to the first 18 participants; however, this issue was corrected by the experimenters thereafter, resulting in a sufficient sample size for evaluating the manipulation.

4. The use of power (1 – β) = .80 is based in convention and corresponds to Cohen’s (1988) argument that the ratio of type 2 to type 1 error should be 4:1, meaning that if α = .05, we should set β = .20, resulting in power (1 – β) = .80.

5. A GPower analysis (Erdfelder et al., 1996) run for a repeated measures ANOVA design at three time points using settings otherwise identical to our initial power analysis reported for our analysis of the relationship at two time points (power (1 – β) set at 0.80, α = .05, two-tailed, and f = .15) shows that a sample size of 108 is needed to detect an effect of the tested size. As such, our sample size (n = 116) is exceeds the required 108, providing sufficient power.

Appendix A:

Descriptive Statistics

Table

	Mean Attitudes Differentiated by Measurement Point and Topic (M, SD)
	Abortion	Death Penalty	Immigration	Healthcare
Pre-Exposure Attitudes (t1)	−20.63 (33.20)	−2.49 (28.55)	9.34 (26.31)	−9.89 (26.25)
Post-Exposure Attitudes (t2)	−13.60 (34.32)	5.22 (29.10)	14.27 (26.82)	−5.99 (25.47)
Delayed Attitudes (t3)	−15.42 (34.65)	4.99 (28.79)	12.36 (27.67)	−7.27 (25.04)
n included, t1	169	166	168	168
n included, t2	171	171	171	169
n included, t3	114	115	116	114

Table

	Immediate Attitude Change (t2-t1) by Format (M, SD)
	Abortion	Death Penalty	Immigration	Healthcare
Signaling/Fact	9.49 (24.85)	11.08 (18.35)	6.06 (16.47)	5.48 (17.62)
Signaling/Fiction	4.22 (19.69)	6.85 (18.32)	3.39 (13.76)	9.26 (18.43)
No Signaling/Fact	5.41 (20.26)	7.35 (21.02)	7.12 (14.70)	.65 (21.78)
No Signaling/Fiction	9.20 (15.64)	8.83 (12.12)	5.64 (18.11)	1.53 (20.52)
n, Signaling/Fact	41	41	41	41
n, Signaling/Fiction	40	39	39	39
n, No Signaling/Fact	43	41	43	43
n, No Signaling/Fiction	45	45	45	44
	Persistent Attitude Change (t3-t1) by Format (M, SD)
	Abortion	Death Penalty	Immigration	Healthcare
Signaling/Fact	7.60 (22.71)	4.74 (22.76)	−.59 (15.58)	.93 (20.14)
Signaling/Fiction	1.52 (19.90)	4.59 (16.66)	2.72 (15.28)	9.15 (20.53)
No Signaling/Fact	6.03 (29.83)	3.34 (16.46)	2.69 (15.73)	5.96 (15.01)
No Signaling/Fiction	6.83 (12.96)	5.85 (16.00)	2.43 (16.41)	−4.96 (23.67)
n, Signaling/Fact	29	29	29	29
n, Signaling/Fiction	28	28	28	28
n, No Signaling/Fact	28	26	28	28
n, No Signaling/Fiction	28	28	28	28

Appendix B:

Correlation Matrix

Summary of Correlations Between Variables (N = 171)

Table

Variable	1	2	3	4	5	6	7	8	9
1. Signal vs. No Signal	-
2. Fact vs. Fiction	−.02
3. Attitude Change (t2-t1)	.05	−.03
4. Attitude Change (t3-t1)^i	−.02	−.05	.66**
5. Counterarguing	.07	.00	−.17*	−.17
6. FF Manip. Check (all texts by condition)	.10	.30**	.02	.04	.27**
7. PI Manip. Check (posttest; signaling texts)^ii	−.12	.07	.02	.11	.04	.09
8. PI Manip. Check (posttest; no signaling texts)^ii	−.04	.10	.20	.16	−.03	.03	.58**
9. Exposure Time	.03	.04	.05	.00	−.09	.04	−.10	.05	–
M	.48	.50	6.21	3.59	3.48	−.40	5.65	5.32	154.15
SD	.50	.50	10.34	9.40	.72	2.72	1.22	1.17	84.46

Notes: *p < .05, **p < .01, ⁱn = 114, ⁱⁱn = 98.