Comparing the effects of simple and refutational narratives in misinformation correction: The moderating roles of correction placement and issue involvement

Abstract

The study examines whether adding a refutational ending to narrative messages improves correction effectiveness and how the effect differs depending on whether the correction message is presented before or after exposure to misinformation. A 2 (narrative format: simple vs refutational narrative) × 2 (correction placement: prebunking vs debunking) between-subjects online experiment (N = 281) with US participants was conducted to correct misinformation about human papilloma virus vaccines. The results suggested that the refutational narrative was more effective in reducing misbeliefs in prebunking, whereas the simple narrative was more effective in debunking. This interaction was further moderated by issue involvement. Theoretical and practical implications are discussed.

Keywords

correction placement HPV vaccine issue involvement misinformation correction narrative persuasion

Amid the growing concerns about rampant health misinformation, scholars have paid increasing attention to message strategies for misinformation correction (Bode and Vraga, 2015; Huang and Wang, 2022; Walter et al., 2021). Although meta-analyses have generally supported the effectiveness of campaign messages debunking misinformation (Chan et al., 2017; Walter et al., 2021; Walter and Murphy, 2018), the accumulated empirical evidence has also revealed the “continued influence” or “persistent effects” of misinformation (Chan et al., 2017). That is, individuals continue to rely on the mental models created by the misinformation even after misinformation is retracted (Thorson, 2016). The lingering effects of misinformation about vaccines on vaccine hesitancy epitomize this harmful impact (Shelby and Ernst, 2013). Given these challenges, scholars have proposed to use “enhanced correctives” such as narrative messages, emotional appeals, and accounts containing causal reasoning (Cappella et al., 2016; Sangalang et al., 2019).

Narratives are considered a powerful tool to combat misinformation as they are especially effective in reducing counterarguing and encouraging emotional responses (Dal Cin et al., 2004). Individuals strongly influenced by misinformation tend to regard correction messages, particularly didactic messages with a strong persuasive intent, as counter-attitudinal attacks and immediately reject them. Narratives, however, may influence those individuals more implicitly and replace the mental models created based on misinformation with a compelling alternative explanation (Sangalang et al., 2019; Wang and Huang, 2021).

To develop enhanced forms of narrative correctives, we draw on the competitive framing and inoculation literature, and experiment on a particular type of narrative by adding a refutational ending. Inoculation research has reported the effectiveness of refutational statements in resisting the influence of subsequent counter-messages (Banas and Rains, 2010). However, the influence of refutational statements has often been tested with didactic messages rather than narratives. Combining narrative persuasion and inoculation literature, this study investigates whether adding a refutational ending can increase narrative effects in correcting misinformation.

In addition to comparing simple and refutational narratives, we are interested in whether their effects differ based on the presentation order of the correction and misinformation messages. Building on the primacy and recency effect literature (Crano, 1977; Haugtvedt and Wegener, 1994), we explore whether it is more effective to provide the correction prior to (i.e. prebunking) or following misinformation exposure (i.e. debunking). The effects of prebunking (e.g. Niederdeppe et al., 2015) and debunking messages (e.g. Bode and Vraga, 2015; Huang and Wang, 2022) have been frequently, but separately, examined in the past. Their effectiveness has rarely been compared and considered as a qualifier for narrative correction (Vraga et al., 2020).

We also examine the moderating role of issue involvement. Defined as perceived personal importance of the discussed topic to individuals, issue involvement often determines the amount of cognitive effort and elaboration devoted to message processing. It likely influences how people react to correction messages and the level of attention they pay to different message features and presentation order.

Specifically, this study examines the effects of narrative format, correction placement, and issue involvement in reducing misbeliefs about human papilloma virus (HPV). Although medical professionals agree that HPV vaccines are an effective measure for cancer prevention, safety concerns, perceived lack of necessity, and inadequate knowledge have led to significant vaccine hesitancy among young adults, teenagers and their parents (John Hopkins Medicine, 2018). HPV vaccination rates remain low and public support for vaccination policies is struggling (University of Chicago Medicine, 2019). In this study, we correct the widespread misinformation that males do not need HPV vaccines as they would not develop cervical cancer. Despite the relatively high awareness of the potential benefits of HPV vaccination for females, fewer people know that HPV vaccines can prevent a common form of throat cancer among males that accounts for 80% of the diagnosed cases. This type of cancer has overtaken cervical cancer and become the leading cancer caused by HPV (Michigan Medicine, 2021). Therefore, it is of practical importance to raise public awareness of the benefits of HPV vaccines for men.

1. Literature review

Challenges in misinformation correction and narrative effects

Researchers have proposed several theoretical accounts to explain why it is challenging to correct misinformation once it has been accepted. Ecker et al. (2011) suggested that individuals tend to build mental models to understand the causal relations among events and are often reluctant to dismiss the presumed cause in established models. Thus, simple rebuttals that do not offer a powerful alternative explanation may fail to correct mental models established based on misinformation (Hamby et al., 2020). Moreover, individuals may engage in motivated reasoning, a tendency to protect their pre-existing attitudes in the face of counter-attitudinal information (Scheufele and Krause, 2019). When individuals have been misinformed, they may regard the correction message as a persuasive attack and are motivated to discredit it, thus reducing the efficacy of correction effort.

These challenges point to the potential of narrative correction. A narrative refers to a message representing “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” (Kreuter et al., 2007: 222). Past research has revealed several mechanisms of narrative persuasion that may make it particularly useful in combating misinformation. First, narratives are the fundamental ways in which individuals make sense of the world surrounding them. By connecting the actions and outcomes experienced by a character in a causal manner, narratives can offer a powerful mental model for the interrelated events and underlying causal explanations (Dahlstrom, 2010). Narrative transportation (Green and Brock, 2000), a state in which individuals are immersed into the story world and picture themselves being part of the story events, is a process wherein individuals mentally simulate the alternative mental model presented in narratives (Busselle and Bilandzic, 2009). Thus, narrative correction may effectively replace the erroneous mental model resulting from misinformation exposure.

Second, narratives can reduce audience resistance to persuasion (Dal Cin et al., 2004; Ratcliff and Sun, 2020). As the core persuasive message is often implied through story events, instead of being explicitly expressed, narratives are less likely to be flagged as persuasive attacks and prompt reactance. More importantly, narrative transportation can occupy cognitive resources and reduce the chance of a critical evaluation of the presented information (Green and Brock, 2000). Hence, the tendency to argue against a correction message may be minimized when it is in a narrative format. Finally, unlike didactic information, narratives are often easy to process and emotionally engaging (Chang, 2013), which can promote the understanding and retention of a correction message.

Despite the promising prospects, empirical evidence of narrative correction has been mixed. Ecker et al. (2020), for example, did not find narrative correction was more effective than non-narrative correction. Sangalang et al. (2019), however, found that narrative correction could effectively reduce misinformed beliefs toward natural tobacco products compared with a no-correction control. The mixed results may suggest that the benefits of using narratives to correct misinformation are moderated by message and presentation features. For example, Huang and Wang (2022) found algorithm-generated narrative correction better activated transportation than narrative correction presented by social media friends. Another study revealed that simple one-sided narrative correction was more effective than two-sided narrative correction for audiences with limited prior experience (Wang and Huang, 2021). Narrative correction with negative emotional endings was also found more effective than simple narratives (Sangalang et al., 2019). Collectively, these empirical studies related to message source, message sidedness and emotional appeals suggest that the efficacy of narrative correction is conditioned by how the narrative is drafted and presented. To optimize the use of narrative correction, research should focus on the message features that potentially enhance or hinder narrative effects. Particularly, this study tests whether a narrative containing ending statements that directly refute misinformation is more effective than a simple narrative without such statements (referred to as refutational narratives vs simple narratives hereafter).

Refutational narratives

Ecker et al. (2020) argued that one potential drawback of narrative correction is that processing story elements and details might distract people from the core message of the correction. Given that the persuasive message of narratives is often implied through the connection of story events, rather than being explicitly stated (Dahlstrom, 2010), individuals need to make additional cognitive efforts to infer the intended message of narrative correction. Focusing on the details of a story during immersion processes can further impede inferential reasoning and reduce processing of the intended message (Slater and Rouner, 2002). To address this limitation, adding a refutational ending that reiterates the intended correction message might increase the ability of narratives to correct misinformation.

The effectiveness of a refutational ending is rooted in the literature highlighting the advantages and disadvantages of explicit persuasive appeals. It is not uncommon to combine explicit persuasive messages with narratives. Epilogues following entertainment education programs are typical examples. Theoretically, these explicit persuasive appeals can bolster the effects of narratives by reinforcing the messages embedded in the narratives. Such reinforcement provides an opportunity for people to rehearse the events and experience the causations of the interrelated events, which increases the retention of the intended message (Moyer-Gusé et al., 2012). Cohen (2016), for example, provided preliminary evidence that adding a supplemental conclusion to a narrative message did help those with a strong eudaimonia motivation (i.e. the motivation to seek meaning from media content) generate more message-consistent beliefs. However, the presence of explicit persuasive appeals like a refutational ending may also increase counterarguing and reactance (Frazer et al., 2021; Moyer-Gusé et al., 2012). The strengths of narratives lie in their ability to facilitate attitude change in subtle ways. However, the use of explicit persuasive appeals may undermine these strengths. As prior research presents conflicting predictions about the effect of a refutational ending in narrative correction, we propose the following research question:

RQ1. Will refutational narratives be more effective than simple narratives in reducing misbeliefs and encouraging behavioral intentions to act on the correction message?

Correction placement

Correction placement is about whether correction messages are placed before or after misinformation exposure. Persuasion research on order effects provides the theoretical underpinnings to discuss the prebunking (correction before misinformation exposure) and debunking (correction after misinformation exposure) effects of correction messages. Researchers have found when people encounter conflicting information, message order can influence their positions (Crano, 1977; Haugtvedt and Wegener, 1994). If they are more influenced by the first message, a primacy effect has occurred. If they are more influenced by the second message, a recency effect has occurred.

Previous research generally concludes that when people are highly motivated to process information, such as when personal relevance is high, a primacy effect is more likely to occur. Haugtvedt and Wegener (1994) argued that individuals with high motivation would scrutinize the first message carefully and form strong attitudes. Under the influence of the strong attitudes, the second communication would be less influential as people merely integrate it into their existing knowledge structure formed when processing the first message. The inoculation theory exemplifies such an effect. It suggests that forewarning messages can strengthen existing attitudes and inoculate individuals against the influence of subsequent counter-opinions (McGuire, 1961). Notably, forewarning is more effective among people who are motivated to protect their attitudes. Empirical evidence has demonstrated that initial forewarnings can induce strong favorable attitudes (Haugtvedt and Wegener, 1994) and elicit reactance to subsequent counter-messages (Niederdeppe et al., 2015). In the battle against misinformation, prebunking messages can not only forewarn people of the existence of misinformation and arguments against it (Pfau et al., 1997), but also show people how misinformation can manipulate them and threaten their freedom to decide (Niederdeppe et al., 2015).

However, a substantial body of persuasion literature also reveals that when people are paying little attention or lack the motivation to process messages, a recency effect is more likely to occur (Crano, 1977; Haugtvedt and Wegener, 1994). Haugtvedt and Wegener (1994) added that the recency effect normally relies on a better recall of the content of the second message. This prediction is supported by recent empirical evidence that fact-focused corrections successfully reduced misperceptions if they were provided after exposure to misinformation but not preceding it (Vraga et al., 2020). A study in crisis management also showed that debunking the negative news coverage an organization previously received successfully restored its image, indicating support for the greater influence of the second communication than the first negative communication (Han et al., 2021). Adding on this evidence, a recent meta-analysis suggested that debunking is generally more effective than prebunking in misinformation correction (Walter and Murphy, 2018). Concurring with the aggregate evidence, we propose that:

H1. Correction placement will have a main effect on persuasion outcomes, such that the debunking strategy will result in more favorable persuasion outcomes than the prebunking strategy, including (a) reducing the targeted misbelief, (b) reducing general misbeliefs, and (c) increasing the behavioral intentions to act on the correction message.

Enhanced effects of refutational narratives as a prebunking strategy

The motivation to process determines when a primacy or recency effect would occur (Crano, 1977; Haugtvedt and Wegener, 1994). People normally have a low motivation to process conflicting information and thus are more subject to a recency effect, making a debunking strategy more effective (Walter and Murphy, 2018). However, it is noticeable that several message features, such as topic controversy and the presence of conflicting views, can increase the motivation to process (Haugtvedt and Wegener, 1994; Sülflow et al., 2019).

Compared with simple narratives, the addition of refutational statements which introduce the existence of counterarguments may enhance message processing and elaboration as the presence of conflicting arguments reduces people’s confidence in making a quick judgment and requires greater cognitive effort to process. As such, refutational statements placed in the first communication may likely increase cognitive processing and reduce the recency effect. According to prior studies, a primacy effect will occur only under the influence of strong attitudes formed when processing the first message (Crano, 1977; Haugtvedt and Wegener, 1994). Thus, when used as a prebunking strategy, refutational narratives will be more effective than simple narratives as they are more likely to elicit the primacy effect, and the influence of misinformation encountered in subsequent communication will be reduced. This prediction is consistent with the inoculation literature, which demonstrates that refutational statements used in a prebunking message are useful to create a structure of reasoning for the audience (Banas and Rains, 2010; Niederdeppe et al., 2015).

However, the advantage of narratives in debunking misinformation lies with their implicit intent, which reduces reactance among people who are influenced by the erroneous mental models established during prior misinformation exposure (Dal Cin et al., 2004; Ratcliff and Sun, 2020). The presence of refutational statements, however, makes the persuasive intent explicit and could potentially elevate audience reactance (Moyer-Gusé et al., 2012). Thus, when used as a debunking strategy, simple narratives without explicit refutational statements may be more effective than refutational narratives. We hypothesize an interaction effect between narrative format and correction placement as follows:

H2. There will be an interaction effect between narrative format and correction placement on persuasion outcomes, such that the refutational narrative will be more effective than the simple narrative when used as a prebunking message, and the simple narrative will be more effective than the refutational narrative when used as a debunking message.

The moderating role of issue involvement

As a variable integral to the dual processing models, issue involvement refers to the perception that the issue discussed is relevant or important to one’s self-concept, goals, and values (Petty and Cacioppo, 1986). When people are highly involved, they scrutinize the information presented to them and argument quality plays a prominent role in determining persuasion outcomes. When people are less involved, they are more influenced by heuristic cues (Petty and Cacioppo, 1986). Although some scholars believe that narrative processing is distinct from the processing of non-narrative information (e.g. Green and Brock, 2000; Slater and Rouner, 2002), recent empirical evidence demonstrates that issue involvement may also condition audience responses to narratives (e.g. Liu and Yang, 2020; Wang and Huang, 2021).

We previously argued that refutational narratives, as a prebunking strategy, are more likely to overcome the recency effect of misinformation than simple narratives as they require more elaboration and effort to synthesize the arguments (Katz et al., 2018). However, not everyone will accept the cognitive demand for processing refutational narratives. Those who are more involved in the issue may devote more effort to message processing. Thus, they may find refutational narratives a cognitive fit to their processing levels. On the other hand, those who are less involved are less likely to elaborate on correction messages. They may find refutational narratives that contain conflicting arguments too demanding to process. In other words, the advantage of a refutational narrative over a simple narrative in prebunking would be more pronounced among the highly involved than lowly involved individuals.

In a debunking scenario, we argued that refutational narratives with a more explicit persuasive intent may be less preferred than simple narratives because they are more likely to elicit audience resistance. Such a difference could also be elevated as people become increasingly involved in the issue. Highly involved individuals are more likely to critically evaluate the debunking message. Hence, the more people are involved, the greater reactance they may generate toward refutational narratives. Therefore, the advantage of a simple narrative over a refutational narrative in debunking will increase as issue involvement increases. We thus propose the following hypothesis:

H3. The interaction effects of narrative format and correction placement on persuasion outcomes, including (a) the targeted misbelief, (b) general misbeliefs, and (c) behavioral intentions, will be more pronounced as levels of issue involvement increase.

2. Method

Design and participants

A 2 (Correction narrative format: refutational narrative vs simple narrative) × 2 (Correction placement: prebunking vs debunking) between-subjects online experiment was conducted to test the proposed hypotheses and research questions. US participants were recruited from Amazon Mechanical Turk. To ensure data quality, three attention-check questions were embedded in the questionnaire, which asked participants to select a specific option as a response (e.g. “agree”). Those who failed any of the attention checks were filtered out. Finally, 281 participants completed the study.

Of the 281 participants (M_age = 38.50, SD_age = 10.36, range = 23–69, males 55.9%), 74.0% identified themselves as Caucasians, followed by 14.9% African Americans, 4.6% Hispanics/Latinos, 3.2% Asians/Pacific Islanders, 2.1% Native Americans/American Indians, and 1.1% multiracial or others. About 18.8% of the participants graduated from high school or received some college education; 60.9% had a 4-year college degree; 20.3% had a postgraduate degree. In our sample, 69.4% of participants reported having at least one child and participants’ prior attitude toward the HPV vaccine was toward the positive end (M = 5.35, SD = 1.45).

Stimuli and experimental treatments

The experimental stimuli included screenshots of four simulated tweets: two of them contained misinformation about HPV vaccines and the other two were correctives. Twitter was selected due to its prominence in spreading health content compared with other social media such as Facebook and Instagram, and the role it plays in disseminating health misinformation such as in the COVID-19 infodemic, especially its higher ratio of low-credibility information than other social media platforms (Yang et al., 2021). The misinformation was tweeted by a fictitious account called Vaccine Choice. The tweet read, “HPV vaccine for women but not men. Only women get HPV infections. Men don’t get cervical cancer.” The corrective was posted by another fictitious account called Health Society, which told a story about Scott Courville’s experience of battling against tonsil cancer caused by an HPV infection. Scott reflected that things could have been different if HPV vaccines were available when he was young. The misinformation and correctives were developed based on existing social media posts, campaign materials, and news articles. The corrective with a refutational ending was 234 words and the one without was 188 words. Due to the word limit of each tweet, the correctives were presented in the form of a Twitter thread. The layout and social media metrics (e.g. the number of likes, comments, and sharing, time stamps, etc.) of the misinformation tweets and correction tweets were generally the same except for the experimental manipulations (see supplemental material).

The correction narrative format was manipulated by varying in presenting or not presenting what misinformation opponents advocated and whether or not the correction message explicitly refuted the misinformation that it intended to correct. In particular, at the end of the refutational narrative, the message revealed that anti-vaccine activists misinformed the public that men do not get cervical cancer and thus HPV vaccines are not important. Then the story character highlighted his own experience that HPV could cause cancer among men, refuting the misinformation that devalued HPV vaccines. By comparison, the simple narrative condition did not explicitly present and refute the misinformation.

The correction placement was operationalized by the different orders in which the misinformation and correction tweets were presented. To imitate the experience of incidental information exposure during social media browsing, two irrelevant tweets about artificial intelligence were provided in between the exposure to the misinformation tweet and the corrective. In the prebunking condition, participants viewed the correction message first and then the irrelevant tweets, followed by the misinformation tweet. In the debunking condition, participants viewed the misinformation and then the two irrelevant tweets; they read the correction message at the end (see supplemental material for details).

Procedure

Upon providing their consent, participants answered a few questions assessing their attitudes toward 10 health issues including e-cigarettes, genetically modified food, opioid use, assisted suicide, medical marijuana, and so on. The question measuring their prior attitude toward HPV vaccines was masked by these issues. Then, the participants were told that their task was to rate a few tweets. They were then randomly assigned to one of the four experimental conditions. Depending on the condition, they viewed the misinformation, correction message, and irrelevant tweets in different orders. After viewing each tweet, they responded to four questions about its credibility. They did so for the two irrelevant tweets as well to reduce sensitization. Following their exposure to the experimental stimuli, we measured misbeliefs, and behavioral intentions about HPV vaccines along with demographic and control variables. Finally, they were debriefed and compensated. The whole process took about 15–20 minutes.

Measures

Targeted misbelief was measured on a 7-point Likert-type scale. Participants were asked how much they agreed that “it is only necessary for girls to be vaccinated against HPV” (M = 4.42, SD = 2.04).

General misbeliefs were measured by asking participants to rate their agreement with five common misbeliefs about HPV and HPV vaccines (Berkowitz, 2015) on a 7-point Likert-type scale. These beliefs include: “The HPV vaccine has terrible side effects,” “There is a cure for HPV,” “You must have sexual intercourse to get HPV,” “The HPV vaccine protects people for life,” and “The HPV vaccine causes teen and preteens to become sexually active” (Cronbach’s α = .82, M = 4.58, SD = 1.34).

Behavioral intentions were measured on a scale from 1 = extremely unlikely to 7 = extremely likely. Participants indicated how likely it was that they would vaccinate their children against HPV if they were at the recommended age (M = 5.88, SD = 1.10).

Issue involvement was measured using five items on a 7-point semantic-differential scale (Zaichkowsky, 1994). Participants indicated how unimportant/important, irrelevant/relevant, worthless/valuable, uninvolving/involving, and not needed/needed they felt about the health topic discussed in the tweets (Cronbach’s α = .92, M = 5.62, SD = 1.06).

Demographic information such as age, gender, race, and education were collected. We also asked whether participants had at least one child, and their prior attitude toward HPV vaccine (1 = extremely negative, 7 = extremely positive). We did not include these variables in the models. We collected such information to understand the sample.

3. Results

Effects of narrative format and correction placement

To answer RQ1 and test H1 and H2, we ran analysis of variance (ANOVA) tests. The results did not reveal a main effect of narrative format on the targeted misbelief, general misbeliefs, and behavioral intentions (RQ1, see Table 1). There was a main effect of correction placement on the targeted misbelief (F(1, 277) = 4.03, p < .05). The participants reported greater targeted misbelief in the prebunking (M = 4.66, SD = 1.98) than debunking condition (M = 4.20, SD = 2.08), suggesting that the debunking strategy was more effective. However, correction placement showed no main effect on general misbeliefs and behavioral intentions. Thus, H1 was partially supported.

Table 1.

Interactions of narrative format and correction placement.

	Targeted misbeliefMean (Standard Deviation)			General misbeliefMean (Standard Deviation)			Behavioral intentionMean (Standard Deviation)
	Prebunk	Debunk	Total	Prebunk	Debunk	Total	Prebunk	Debunk	Total
Simple narrative	4.82 (1.86)	3.94 (2.08)	4.36 (2.02)	4.76 (1.42)	4.40 (1.28)	4.57 (1.35)	5.68 (1.11)	5.97 (1.05)	5.83 (1.09)
Refutational narrative	4.52 (2.08)	4.43 (2.07)	4.47 (2.07)	4.43 (1.30)	4.72 (1.36)	4.58 (1.33)	5.87 (1.10)	5.95 (1.14)	5.91 (1.12)
Total	4.66 (1.98)	4.20 (2.08)		4.58 (1.36)	4.57 (1.32)		5.78 (1.10)	5.96 (1.09)
	F	p		F	p		F	p
Message format	0.14	.71		0.00	.96		0.44	.51
Correction placement	4.03	<.05		0.04	.83		1.96	.16
Format × placement	2.59	.11		4.18	<.05		1.64	.20

There was a significant interaction effect on general misbeliefs (F(1, 277) = 4.18, p < .05). The results suggest that the participants reported fewer general misbeliefs in the prebunking refutational narrative condition than in the prebunking simple narrative condition. In contrast, they reported more general misbeliefs in the debunking refutational narrative condition than in the debunking simple narrative condition. There was no interaction effect on the targeted misbelief and behavioral intentions (see Figure 1). Thus, H2 was partially supported.

Figure 1.

Interaction effects of narrative format and correction placement on general misbeliefs.

Moderating effect of issue involvement

To examine the moderating effect of issue involvement (H3), we ran PROCESS Model 3 with 5000 bootstrapped samples and 95% bias adjusted confidence intervals (Hayes, 2018). After considering the moderating effects of issue involvement, the interaction effect of narrative format and correction placement remained on the targeted misbelief (B = −6.67, SE = 2.67, p < .05, [−11.91, −1.42]) and general beliefs (marginally significant, B = −3.20, SE = 1.75, p = .07, [−6.65, 0.25]), again generally confirming the results for H2. There was also a significant two-way interaction of correction placement and involvement on the targeted misbelief (B = −1.69, SE = 0.75, p < .05, [−3.17, −0.21]). However, these results should be interpreted in light of the significant three-way interaction effect on the targeted misbelief (B = 1.33, SE = 0.46, p < .01, [0.42, 2.25]) and general misbeliefs (B = 0.69, SE = 0.31, p < .05, [0.09, 1.30]) (see the cell means in Table 2 and three-way interaction effect in Figures 2 and 3). When used as a prebunking strategy, the advantage of the refutational narrative message over the simple narrative message in reducing the targeted misbelief was more pronounced as the levels of issue involvement increased. When used as a debunking strategy, the advantages of the simple narrative message over the refutational narrative message in reducing the targeted misbelief and general misbeliefs became more pronounced as the levels of issue involvement increased. The three-way interaction effect was, however, non-significant on behavioral intentions.

Table 2.

Effects of narrative format, correction placement, and issue involvement.

	Targeted misbelief		General misbeliefs		Behavioral intentions
	Coeff	p	Coeff	p	Coeff	p
Main effects
Narrative format	2.89	.13	0.47	.70	−1.81	.07
Correction placement	7.84	.07	3.65	.20	−2.81	.22
Issue involvement	0.70	.18	0.19	.58	−0.25	.36
Two-way interaction effects
Format × placement	−6.67	<.05	−3.20	.07	2.97	.16
Format × involvement	−0.57	.09	−0.14	.58	0.35	<.05
Placement × involvement	−1.69	<.05	−0.84	.09	0.55	.17
Three-way interaction effect
Format × placement × involvement	1.33	<.01	0.69	<.05	−0.37	.14
Conditional effects	b	SE	t	p	LLCI	ULCI
DV: targeted misbelief
Prebunking
Low involvement	0.30	0.49	0.61	.54	−0.67	1.27
Average involvement	−0.30	0.35	−0.85	.40	−0.99	0.39
High involvement	−0.90	0.50	−1.78	.08	−1.89	0.09
Debunking
Low involvement	−0.29	0.50	−0.59	.55	−1.27	0.68
Average involvement	0.51	0.33	1.53	.13	−0.15	1.17
High involvement	1.32	0.46	2.87	<.01	0.42	2.22
DV: general misbeliefs
Prebunking
Low involvement	−0.18	0.32	−0.57	.57	−0.82	0.45
Average involvement	−0.34	0.23	−1.45	.15	−0.79	0.12
High involvement	−0.49	0.33	−1.47	.14	−1.14	0.16
Debunking
Low involvement	−0.21	0.33	−0.66	.51	−0.86	0.43
Average involvement	0.37	0.22	1.67	.10	−0.07	0.80
High involvement	0.95	0.30	3.14	<.01	0.35	1.54
DV: behavioral intentions
Prebunking
Low involvement	−0.20	0.26	−0.78	.44	−0.72	0.31
Average involvement	0.17	0.19	0.91	.36	−0.20	0.54
High involvement	0.54	0.27	2.03	<.05	0.02	1.07
Debunking
Low involvement	0.10	0.26	0.38	.70	−0.42	0.62
Average involvement	0.09	0.18	0.49	.63	−0.26	0.44
High involvement	0.07	0.24	0.30	.77	−0.41	0.55

SE: standard error; LLCI: lower limit confidence interval; ULCI: upper limit confidence interval; DV: dependent variable.

Figure 2.

Three-way interaction on the targeted misbelief.

Figure 3.

Three-way interaction on general misbeliefs.

4. Discussion

This study contributes to the prior literature on narrative correction in four ways. First, we examined whether a message feature, the addition of a refutational ending, improves or hinders the efficacy of narrative correction. We found no main effect of this message feature on misbeliefs and behavioral intentions. Second, we compared the prebunking and debunking strategies in one study. Recent meta-analyses have concluded that the debunking strategy is more effective than the prebunking strategy in correcting the influence of misinformation. This study confirmed that this is generally the case when narratives are used to correct misbeliefs. However, it also depends on how the correction narrative is drafted. Third, we revealed that the effects of the message feature in the narrative correction are conditioned by correction placement. Our findings showed that when used as a prebunking strategy, refutational narratives were more effective than simple narratives in resisting changes in beliefs. When used as a debunking strategy, simple narratives were more effective than refutational narratives in correcting the influence of previous exposure to misinformation. Fourth, we tested the moderating role of issue involvement. We determined that the interaction effects of narrative format and correction placement became more pronounced among participants who were more involved with the issue.

We did not find a main effect of narrative format on correction outcomes including the targeted misbelief, general misbeliefs, and behavioral intentions. However, we did establish that the placement of the correction had a main effect on the targeted misbelief, which confirmed the findings of recent meta-analyses (Walter and Murphy, 2018) that the debunking strategy was more effective than the prebunking strategy in general in countering misbeliefs. To clarify, the targeted misbelief was the misbelief induced by the misinformation while general misbeliefs referred to overall misperceptions about HPV vaccine. Thus, the effect on the targeted misbelief represents the change on a factual belief due to correction while the effect on the general misbeliefs may predict overall perception or attitudinal change toward HPV vaccine. The recency effect may be able to explain the impact of correction placement on the target misbelief. Unless instructed specifically to pay more attention to the messages, people tend to engage in little elaboration of the messages they receive. The second communication due to its recency is more influential than the first communication in determining the belief. Therefore, debunking was more effective in reducing the targeted misbelief than prebunking.

The two-way interaction effect of narrative format and correction placement found in ANOVA and regression models when issue involvement was considered revealed that, in the prebunking condition, refutational narratives were more effective than simple narratives in reducing the targeted misbelief and general misbeliefs. In other words, refutational narratives were more effective in preventing people from changing their beliefs as a result of subsequent exposure to misinformation. The results support the Inoculation Theory about the effect of refutation (Banas and Rains, 2010) and extend its application in argument-based rhetorical messages to immersion-based narrative messages.

The two-way interaction effect also suggested that, in the debunking condition, simple narratives were more effective in countering general misbeliefs than refutational narratives. This pattern may be explained by the continued influence of misinformation (Chan et al., 2017; Thorson, 2016) and the different levels of reactance elicited by simple and refutational narratives (Frazer et al., 2021; Moyer-Gusé et al., 2012). When individuals are exposed to misinformation, they would establish a belief system rooted in the misinformation. Due to the continued and negative impact of misinformation, people are likely to generate reactance to subsequent persuasion which contradicts their beliefs. As such, debunking messages that can reduce audience reactance are advantageous in producing desired correction effects. Prior research has suggested that simple narratives hold the potential to induce less reactance than refutational narratives (Moyer-Gusé et al., 2012). Without explicit refutational statements, simple narratives are drafted to persuade more implicitly. Audiences are thus less likely to detect their persuasive intent and generate psychological reactance (Dal Cin et al., 2004). Consequently, when a debunking strategy is used, simple narratives were more efficacious in correcting misbeliefs than refutational narratives under the continued influence of misinformation.

We also determined that the two-way interaction effect of narrative format and correction placement depended on the participants’ levels of issue involvement. Specifically, the advantage of refutational narratives in prebunking and the advantage of simple narratives in debunking became more pronounced as the participants’ level of issue involvement increased. We concluded that in the prebunking scenario, refutational narratives that require more cognitive elaboration will be more effective in overcoming the recency effect of misinformation. As the dual-process models (e.g. Elaboration Likelihood Model, Petty and Cacioppo, 1986) predict, those who are strongly involved in an issue devote more cognitive effort to processing messages about it. Thus, they find the cognitive demands required to process refutational narratives more acceptable than those who are less involved with the issue. The advantage of refutational narratives over simple narratives will be more pronounced among those who are more involved with the issue. In the debunking condition, simple narratives showed an advantage over refutational narratives as they tend to elicit less reactance. Reactance often takes the form of negative cognitions and emotions (Dillard and Shen, 2005). Greater levels of cognitive elaboration are often needed to generate negative cognitions during message processing. Given the positive association between cognitive elaboration and issue involvement (Petty and Cacioppo, 1986), people who are highly involved in the issue would be likely to generate more reactance to debunking messages than who do not care. As a result, we observed stronger differences between simple and refutational narratives in their debunking effectiveness as the level of issue involvement increased.

Although our findings demonstrated the combinatory effects of narrative type, correction placement and involvement on the targeted misbelief and general misbeliefs, there was an overall lack of effects on behavioral intention. Although beliefs are typically considered an important antecedent to behavioral intention, other factors such as perceptions of efficacy may also influence individuals’ intentions to perform certain behaviors (Garcia and Mann, 2003). While correction messages may reduce misbeliefs and promote factual beliefs, it is important to address other factors to motivate people to move from “I believe” to “I will.”

Our findings also offer some useful suggestions for campaign efforts to correct misinformation. Given the increasing use of storytelling in correction messages, our findings suggest that how story messages are drafted and placed also matter. Adding a refutational ending to stories can make the core correction argument more prominent and help audiences more easily grasp the persuasive subtext underlying the story events. However, the benefits of this persuasive strategy are conditional. Adding a refutational ending can be useful in forewarning audiences of the possible misleading statements they might encounter and better prepare them to resist their influence. However, this strategy may backfire when audiences have already been exposed to the misinformation. To debunk a piece of misinformation, a simple narrative message without refutational statements may be more effective in reducing false beliefs generated by the misinformation and reducing negative misperceptions.

Notably, a possible constraint for placing different types of correction narratives is that campaigners often cannot fully control the order in which audiences view misinformation and correction messages. Targeting the right audience with appropriate messages involves carefully designed strategies and more advanced algorithms. Some existing options of behavioral targeting on social media can be useful. For example, different correction narratives could be presented depending on whether audiences have browsed certain websites or topics related to misinformation. A debunking strategy could be used among audiences with such browsing histories. For widespread misinformation, simple narratives may be more generally applicable options. Also, our findings call for social media platforms to develop more advanced targeting and message display algorithms. These would be valuable tools for health organizations to deliver correction messages to the public with greater precision and effectiveness.

Moreover, it is valuable for campaigners to consider the audience’s level of issue involvement when disseminating their correction messages. Those who are highly involved in the issue may be more motivated to process correction messages. Thus, they may pay more attention to different message features and arrangements, and these differences may influence them more. Understanding this point is particularly important when conducting online campaigns. The targeting functions of digital platforms can be used to select audiences with particular interests. Campaigners can use related variables in these categories (e.g. interest in health issues) as proxies for issue involvement and select the best campaign strategy.

Limitations and future studies

Despite its contribution to prior literature and practical knowledge, the study also has several limitations that suggest potential directions for future research. First, the US sample was recruited from the general population while the issue is most relevant to people who face an HPV decision for themselves or offspring. We made this decision given our interest in understanding the role played by participants’ self-reported levels of issue involvement. We believe we could reach a better spectrum of individuals with different levels of issue involvement from the general population than from a more targeted population. Questions, however, arise about whether behavioral relevance indicators such as if people have age-specific children facing the HPV vaccination decision would report greater involvement. In our study, we found that issue involvement was weakly but significantly correlated with whether participants have children (r = .12, p < .05) and participants’ prior attitude toward HPV vaccines (r = .20, p < .001). We acknowledge that the current sample reported a slightly more favorable prior attitude toward HPV vaccine, possibly due to the sample characteristics. Future research can focus on a more targeted sample by recruiting only young adults and parents of age-relevant children who face the decision for a better understanding of their belief changes and actual vaccination decisions, which will also help draw real-world public health implications. Meanwhile, the focus on the single issue of HPV vaccination may limit the generalizability of the findings of the current study. Future research should replicate the study design in other health contexts and see if similar patterns emerge.

Second, for better experimental control so that the participants would not be distracted by other information such as ads, number of likes, and comments, we used screenshots of social media posts in our testing. We tried to mimic the experience of incidental information exposure on social media by including irrelevant posts in between the misinformation and correction message. However, we are aware that this experience was still somewhat artificial. It is unclear whether our findings would hold in a real social media setting, where individuals are exposed to a greater amount of information and other interface cues. It would be beneficial for future investigations to incorporate field data to validate this study’s findings.

In addition, the sample size may be relatively small to fully probe detailed patterns of the three-way interaction effect between narrative format, correction placement, and issue involvement. To ensure a robust test of the three-way interaction effect, we employed a linear multiple regression approach using PROCESS instead of converting issue involvement into a categorical variable. Post hoc power analysis with G*Power 3.1 suggested that for a multiple regression analysis at p < .05, the sample provided a power greater than .99 to detect medium (f² = .15) to large (f² = .35) effect sizes (Cohen, 1988), although the power might be limited (.35) when effect sizes were small at f² = .02. Despite the linear approach in inferential tests, categorizing issue involvement is a necessary step to probe and visualize the interactions. Thus, we encourage readers to interpret the specific patterns with caution.

Also, behavioral intentions were not measured with a specific time frame as suggested in the reasoned action research. This may explain the inconsistencies with findings in the targeted and general misbeliefs. We did not measure reactance and officially examine this psychological mechanism. The experiment involves exposure to two different messages (i.e. the misinformation and correction message) in the same experiment session. This creates difficulty for appropriately measuring reactance to the correction message. Measuring reactance following the exposure to both messages may require additional instructions to specify the message we are referring to for the measure. To do so, we may need to repeat the message content and thus double the dosage of manipulation. If reactance is measured immediately following correction exposure, it may also unnecessarily sensitize participants and make the message experience even more artificial. Nonetheless, formal testing reactance as a mediator is necessary in future research to corroborate the proposed mechanism. Perhaps this can be more appropriately done in a two-wave experimental design with days instead of minutes apart. With a longer delay between misinformation and correction exposure, it will also become possible to examine counterarguing or reactance to each message without unnecessarily sensitizing participants. Measuring counterarguing against or reactance to misinformation and correction messages will enrich our understanding of the underlying psychological mechanisms of misinformation correction. Replicating the patterns with a different set of messages and samples will also help make sure the effects are not limited just to this topic or sample.

Supplemental Material

sj-docx-1-pus-10.1177_09636625231168995 – Supplemental material for Comparing the effects of simple and refutational narratives in misinformation correction: The moderating roles of correction placement and issue involvement

Supplemental material, sj-docx-1-pus-10.1177_09636625231168995 for Comparing the effects of simple and refutational narratives in misinformation correction: The moderating roles of correction placement and issue involvement by Weirui Wang and Yan Huang in Public Understanding of Science

Footnotes

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was supported by the Arthur W. Page Center for integrity in public communication (grant number 1920NPC005).

ORCID iDs

Weirui Wang

Yan Huang

Supplemental material

Supplemental material for this article is available online.

Author biographies

Weirui Wang is an Associate Professor at the Department of Communication, Florida International University. She holds a PhD in mass communication from the Pennsylvania State University. Her research focuses on health and science communication for social changes, with an emphasis on understanding the impact of culture and social media.

Yan Huang is an Assistant Professor of Integrated Strategic Communication at the Jack J. Valenti School of Communication, University of Houston. She holds a PhD in mass communication with a minor in statistics from the Pennsylvania State University. Huang’s research is at the intersection of strategic communication, persuasion, and media psychology. It investigates how content features and technological aspects of media campaigns shape audience responses to health, risk, and prosocial messages.

References

Banas

Rains

(2010) A meta-analysis of research on inoculation theory. Communication Monographs 77(3): 281–311.

Berkowitz

(2015) HPV and cancer: 9 myths busted. Available at: https://www.mdanderson.org/publications/focused-on-health/FOH-HPV-myths.h14-1589835.html

Bode

Vraga

(2015) In related news, that was wrong: The correction of misinformation through related stories functionality in social media. Journal of Communication 65(4): 619–638.

Busselle

Bilandzic

(2009) Measuring narrative engagement. Media Psychology 12(4): 321–347.

Cappella

Maloney

Ophir

Brennan

(2016) Interventions to correct misinformation about tobacco products. Tobacco Regulatory Science 2: 186–197.

Chan

Jones

Hall Jamieson

Albarracín

(2017) Debunking: A meta-analysis of the psychological efficacy of messages countering misinformation. Psychological Science 28(11): 1531–1546.

Chang

(2013) Imagery fluency and narrative advertising effects. Journal of Advertising 42(1): 54–68.

Cohen

(2016) Exploring subtext processing in narrative persuasion: The role of eudaimonic entertainment-use motivation and a supplemental conclusion scene. Communication Quarterly 64(3): 273–297.

Cohen

(1988) Statistical Power Analysis for the Behavioral Sciences. Hillsdale, NJ: Lawrence Erlbaum Associates.

10.

Crano

(1977) Primacy versus recency in retention of information and opinion change. The Journal of Social Psychology 101(1): 87–96.

11.

Dahlstrom

(2010) The role of causality in information acceptance in narratives: An example from science communication. Communication Research 37: 857–875.

12.

Dal Cin

Zanna

Fong

(2004) Narrative persuasion and overcoming resistance. In: Knowles

Linn

(eds) Resistance and Persuasio. Hillsdale, NJ: Lawrence Erlbaum Associates, pp. 175–191. Available at: http://books.google.com/books?hl=en&lr=&id=iTzmK4mxFIcC&oi=fnd&pg=PA175&dq=Narrative+Persuasion+and+Overcoming+Resistance&ots=n5w6E-3K2y&sig=RzQjT89jvCLN4nJPa9GNKI4vj_s

13.

Dillard

Shen

(2005) On the nature of reactance and its role in persuasive health communication. Communication Monographs 72(2): 144–168.

14.

Ecker

UKH

Butler

Hamby

(2020) You don’t have to tell a story! A registered report testing the effectiveness of narrative versus non-narrative misinformation corrections. Cognitive Research: Principles and Implications 5(64): 1–26.

15.

Ecker

UKH

Lewandowsky

Swire

Chang

(2011) Correcting false information in memory: Manipulating the strength of misinformation encoding and its retraction. Psychonomic Bulletin and Review 18(3): 570–578.

16.

Frazer

Robinson

Knobloch-Westerwick

(2021) Narratives’ impacts on attitudes: Do signaling of persuasive intent and fictionality matter? Communication Studies 72(3): 347–365.

17.

Garcia

Mann

(2003) From “I wish” to ‘I will’: Social-cognitive predictors of behavioral intentions. Journal of Health Psychology 8(3): 347–360.

18.

Green

Brock

(2000) The role of transportation in the persuasiveness of public narratives. Journal of Personality and Social Psychology 79(5): 701–721.

19.

Hamby

Ecker

Brinberg

(2020) How stories in memory perpetuate the continued influence of false information. Journal of Consumer Psychology 30(2): 240–259.

20.

Han

Davies

Grimes

(2021) Recency effects in the buffering of negative news by corporate social responsibility advertising. Corporate Communications: An International Journal 26(2): 382–402.

21.

Haugtvedt

Wegener

(1994) Message order effects in persuasion: An attitude strength perspective. Journal of Consumer Research 21(1): 205.

22.

Hayes

(2018) Introduction to Mediation, Moderation, and Conditional Process Analysis: A Regression-Based Approach, 2nd edn. New York, NY: Guilford Publications.

23.

Huang

Wang

(2022) When a story contradicts: Correcting health misinformation on social media through different message formats and mechanisms. Information, Communication & Society 25(8): 1192–1209.

24.

John Hopkins Medicine (2018) The HPV vaccine: Why parents really choose to refuse. Available at: https://www.hopkinsmedicine.org/news/newsroom/news-releases/the-hpv-vaccine-why-parents-really-choose-to-refuse

25.

Katz

Erkkinen

Lindgren

Hatsukami

(2018) Assessing the impact of conflicting health warning information on intentions to use E-cigarettes- An application of the heuristic-systematic model. Journal of Health Communication 23: 874–885.

26.

Kreuter

Green

Cappella

Slater

Wise

Storey

, et al. (2007) Narrative communication in cancer prevention and control: A framework to guide research and application. Annals of Behavioral Medicine 33: 221–235.

27.

Liu

Yang

(2020) The role of temporal distance perception in narrative vs. non-narrative persuasion related to e-cigarettes. Journal of Health Communication 25(7): 543–553.

28.

McGuire

(1961) The effectiveness of supportive and refutational defenses in immunizing and restoring beliefs against persuasion. Sociometry 24(2): 184–197.

29.

Michigan Medicine (2021) Few young adult men have gotten the HPV vaccine. Available at: https://medicine.umich.edu/dept/family-medicine/news/archive/202104/few-young-adult-men-have-gotten-hpv-vaccine

30.

Moyer-Gusé

Jain

Chung

(2012) Reinforcement or reactance? Examining the effect of an explicit persuasive appeal following an entertainment-education narrative. Journal of Communication 62(6): 1010–1027.

31.

Niederdeppe

Heley

Barry

(2015) Inoculation and narrative strategies in competitive framing of three health policy issues. The Journal of Communication 65(5): 838–862.

32.

Petty

Cacioppo

(1986) The elaboration likelihood model of persuasion. In: Petty

Cacioppo

(eds) Communication and Persuasion: Central and Peripheral Routes to Attitude Change. New York, NY: Springer, pp. 1–24.

33.

Pfau

Tusing

Koerner

Lee

Godbold

Penaloza

, et al. (1997) Enriching the inoculation construct: The role of critical components in the process of resistance. Human Communication Research 24(2): 187–215.

34.

Ratcliff

Sun

(2020) Overcoming resistance through narratives: Findings from a meta-analytic review. Human Communication Research 46(4): 412–443.

35.

Sangalang

Ophir

Cappella

(2019) The potential for narrative correctives to combat misinformation. Journal of Communication 69(3): 298–319.

36.

Scheufele

Krause

(2019) Science audiences, misinformation, and fake news. Proceedings of the National Academy of Sciences of the United States of America 116(16): 7662–7669.

37.

Shelby

Ernst

(2013) Story and science: How providers and parents can utilize storytelling to combat anti-vaccine misinformation. Human Vaccines & Immunotherapeutics 9(8): 1795–1801.

38.

Slater

Rouner

(2002) Entertainment-education and elaboration likelihood: Understanding the processing of narrative persuasion. Communication Theory 12(2): 173–191.

39.

Sülflow

Schäfer

Winter

(2019) Selective attention in the news feed: An eye-tracking study on the perception and selection of political news posts on Facebook. New Media & Society 21(1): 168–190.

40.

Thorson

(2016) Belief echoes: The persistent effects of corrected misinformation. Political Communication 33(3): 460–480.

41.

University of Chicago Medicine (2019) Public perception of scientific uncertainty is linked to HPV vaccine policy support. Available at: https://www.uchicagomedicine.org/forefront/cancer-articles/public-perception-of-scientific-uncertainty-is-linked-to-hpv-vaccine-policy-support

42.

Vraga

Kim

Cook

Bode

(2020) Testing the effectiveness of correction placement and type on Instagram. The International Journal of Press/politics 25(4): 632–652.

43.

Walter

Murphy

(2018) How to unring the bell: A meta-analytic approach to correction of misinformation. Communication Monographs 85(3): 423–441.

44.

Walter

Brooks

Saucier

Suresh

Walter

Brooks

, et al. (2021) Evaluating the impact of attempts to correct health misinformation on social media: A meta-analysis. Health Communication 36(13): 1776–1784.

45.

Wang

Huang

(2021) Countering the “harmless e-cigarette” myth: The interplay of message format, message sidedness, and prior experience with e-cigarette use in misinformation correction. Science Communication 43(2): 170–198.

46.

Yang

Pierri

Hui

Axelrod

Torres-Lugo

Bryden

, et al. (2021) The COVID-19 infodemic: Twitter versus Facebook. Big Data & Society. Epub ahead of print 5 May. DOI: 10.1177/20539517211013861.

47.

Zaichkowsky

(1994) The personal involvement inventory: Reduction, revision, and application to advertising. Journal of Advertising 23(5): 59–70.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.71 MB